THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY
2010, 63 (2), 371 – 400
Differential effects of constraints in the processing
of Russian cataphora
Nina Kazanina
Department of Experimental Psychology, University of Bristol, Bristol, UK
Colin Phillips
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Department of Linguistics, Neuroscience and Cognitive Science Program, University of Maryland, College Park, MD, USA
Anaphoric relations between pronouns and their antecedents are subject to a number of different
linguistic constraints, which exclude the possibility of coreference in specific syntactic or discourse contexts. Constraints on anaphora may, in principle, impact online sentence processing in a couple of
different ways. They may act as constraints on the generation of interpretations, preventing illicit anaphoric relations from ever being considered. Alternatively, they may act as later filters on interpretations, rejecting candidate interpretations after initial consideration. A number of previous studies
have sought to determine which of these mechanisms accurately describes the online impact of constraints on anaphora. The current studies present evidence that there is no uniform answer to this
question, and that the two mechanisms are both used, for different constraints. Evidence for this is
drawn from studies on the processing of two constraints on backwards anaphora or cataphora in
Russian that apply in superficially similar contexts but that differ in a number of respects. One
self-paced reading study and two judgement studies are reported. The self-paced reading study
manipulates the gender congruency between a pronoun and a following name in three pairs of conditions. In conditions where the pronoun – name configuration violates no constraints on anaphora
a gender mismatch effect was observed following the name, as in previous studies, suggesting that
comprehenders actively search for an antecedent following a cataphoric pronoun. In conditions
where the pronoun – name configuration violates Principle C of the classical binding theory no
effect of the gender manipulation was observed, suggesting that comprehenders do not even consider
the possibility of interpretations that violate this constraint. In conditions where the pronoun– name
configuration violates a Russian-specific constraint on cataphora a gender match effect was observed
following the name, the reverse of the finding in the no-constraint conditions, suggesting that the
constraint applies as a filter on candidate interpretations.
Keywords: Backwards anaphora; Cataphora; Sentence processing; Russian; Principle C.
Correspondence should be addressed to Nina Kazanina, Department of Experimental Psychology, University of Bristol, 12a,
Priory Road, Bristol, BS8 1TU, UK. E-mail: [email protected]
We are grateful to John Bailyn, Roger van Gompel, Masaya Yoshida, and three anonymous reviewers for valuable feedback on
this research, and to Elena Kislik and Andrey Sechin for their help in running the experiments. We acknowledge the support of the
National Science Foundation (NSF) grant DGE-0801465 to the University of Maryland.
# 2009 The Experimental Psychology Society
http://www.psypress.com/qjep
371
DOI:10.1080/17470210902974120
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
Anaphoric expressions such as pronouns (e.g., she,
him, their) and reflexives (e.g., herself, yourself ) are
expressions that are dependent for their interpretation on other material in the context where they
are used. Anaphora resolution is governed by a
variety of different constraints at the levels of
syntax, semantics, pragmatics, and prosody
(Barss, 2003; Büring, 2005; Huang, 2004), and
these constraints vary across languages and across
different anaphoric items within individual
languages (Cole, Hermon, & Huang, 2000;
Kaiser & Trueswell, 2008; Koster & Reuland,
1992). Consequently, anaphora resolution has
proven to be a fertile testing ground for theories
of language processing (e.g., Garnham, 2001)
and in particular for studies of the time course
of constraint application in language comprehension.
The current study investigates how grammatical constraints impact the search for antecedents
for pronouns, with a focus on backwards anaphora
or cataphora, a type of pronominal dependency in
which a pronoun linearly precedes its antecedent.
(We use the term “anaphora” here as a neutral
term that covers both forward-looking and backward-looking dependencies.) A number of
previous studies have investigated whether
grammatical constraints on anaphora remove
potential antecedents from consideration during
online sentence comprehension. Most studies
have found that grammatical constraints do
impact anaphora resolution, but there are conflicting findings on exactly when and how ungrammatical antecedents are removed from consideration.
Here we examine this issue through the lens of a
pair of constraints on backwards anaphora in
Russian that apply to superficially similar structures yet differ in a number of respects.
Grammatical constraints on anaphora could, in
principle, impact the search for antecedents in at
least two ways. Consider a sentence like John
thinks that Bill resents him, in which the pronoun
him may corefer with the main clause subject
John but not with its clause-mate subject Bill,
due to Principle B of the classical binding theory
(Chomsky, 1981) or a more recent counterpart
(e.g., Reinhart & Reuland, 1993; Reuland,
372
2001). The impact of the constraint may be to constrain the generation of candidate antecedents,
such that the grammatically inaccessible Bill is
never even considered as an antecedent for the
pronoun. Alternatively the constraint may impact
a later processing stage, such that both John and
Bill are initially considered as candidate antecedents, and Bill is subsequently filtered out by the
constraint. Sturt (2003) describes these alternatives as “early” and “late” filters, respectively, but
we prefer to restrict the term “filter” to the latter
cases only.) Sturt (2003) raises a third possibility,
where grammatically inaccessible antecedents are
blocked from consideration at an initial automatic
stage, but under some circumstances may be
reconsidered at a later stage.
Existing studies on the online impact of
binding constraints have yielded conflicting
results. In a study using a cross-modal lexical
decision task Nicol and Swinney argued that
Principles A and B of the classical binding
theory act as constraints on the generation of candidate antecedents (Nicol & Swinney, 1989). For
example, lexical decision times for words presented
immediately after the pronoun him in sentences
like The boxer told the swimmer that the doctor for
the team would blame him for the recent injury
showed facilitation for semantic associates of the
two grammatically accessible antecedents (boxer,
swimmer) but not for associates of the grammatically inaccessible antecedent (doctor). Similar
conclusions about the role of constraints on anaphora have been drawn from studies using selfpaced reading measures (Clifton, Kennison, &
Albrecht, 1997; Lee & Williams, 2006), eye tracking during reading (Sturt, 2003), and event-related
brain potentials (Xiang, Dillon, & Phillips, 2009).
However, another group of studies has found evidence of consideration of grammatically inaccessible antecedents, leading to the conclusion that
constraints on anaphora act as late filters on a
broader initial set of candidates. For example,
Badecker and Straub’s (2002) self-paced reading
results suggest interference from grammatically
inaccessible antecedents in the processing of the
pronoun in “multiple match” sentences like John
thought that Bill owed him another chance to solve
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
the problem, where the noun phrase (NP) Bill is an
impossible antecedent due to Principle B. Similar
conclusions have been drawn from studies using
self-paced reading (Kennison, 2003) and eye
tracking in the visual-world paradigm (Runner,
Sussman, & Tanenhaus, 2006).
The discrepant results from previous studies
may indicate that the parser uses multiple mechanisms to implement constraints on coreference.
However, it is also possible that the discrepant
results from previous studies may partly be due
to methodological differences across studies.
Measures like cross-modal lexical decision (Nicol
& Swinney, 1989) provide good evidence of the
activation of specific candidate antecedents, but
offer limited time course information. Word-byword reading measures such as self-paced reading
(Badecker & Straub, 2002) and eye tracking
(Sturt, 2003) provide more detailed time course
information but less direct measures of antecedent
activation. Even measures such as eye tracking in
the visual-world paradigm, which promise continuous data and more transparent measures of
referent activation, rely on specific assumptions
about the link between gaze patterns and anaphoric dependency formation. Another possible source
of discrepancy across different studies may be
related to how closely the accessible and inaccessible antecedents were matched in discourse prominence, a property that modulates the impact of
potential antecedents on processing (see
Badecker & Straub, 2002; Sturt, 2003 for discussion). In the current study we test whether
multiple mechanisms are required to implement
constraints on coreference by using a withinsubjects comparison of different constraints and
by turning to a different type of anaphoric relation
from what has been considered in most previous
studies.
Most previous studies on binding constraints
in language processing have focused on cases of
forwards anaphora, in which comprehenders
encounter a pronoun or reflexive after encoding the
potential antecedents in memory. Consequently,
forwards anaphora resolution is a retrospective
process that potentially requires multiple candidate
antecedents to be considered simultaneously. Such
situations may be particularly sensitive to the contrasting salience of different antecedents, and this
may mask the effects of grammatical constraints.
The parser’s failure to consider a potential antecedent during online interpretation can only be
attributed to a grammatical constraint if we are
confident that it is not simply ignored due to low
discourse prominence. A straightforward way to
avoid this concern is to turn to cases of backwards
anaphora/cataphora. Cataphora is less frequent
than forwards anaphora, yet both natural and
common in English—for example, When she
enters the classroom, Zoe sits down at the art table.
Although some instances of apparent cataphora
may turn out to be instances of cross-sentential
forwards anaphora, there is ample evidence for
true backwards anaphora. For example, van
Hoek (1997) documents numerous instances of
naturally occurring discourse initial backwards
anaphora. From the perspective of online
interpretation the resolution of backwards anaphora must proceed differently from the resolution
of forwards anaphora. After identifying a cataphoric pronoun a comprehender may consider each
subsequent noun phrase as a potential antecedent
and evaluate it individually at the time of its
maximal salience in the discourse (i.e., when it is
the current incoming word or phrase).
In an eye-tracking study van Gompel and
Liversedge (2003) investigated the time course of
reference resolution between cataphoric pronouns
and grammatically accessible antecedents. Using
sentences like When she was fed up, the {girl/boy}
visited the {boy/girl} very often, they found that
comprehenders attempt to link a cataphoric
pronoun (she) to a potential antecedent (the girl/
boy) before they have confirmed that the
expression is compatible with the pronoun in
gender. Evidence for this comes from a slowdown
in reading times when readers encountered a noun
phrase that mismatched in gender with a preceding pronoun. Van Gompel and Liversedge argue
that this gender mismatch effect reflects an unsuccessful attempt to create a referential dependency
between the name and the pronoun, and further
that this effect could only have arisen if readers
attempt to create referential dependencies before
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
373
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
verifying their semantic well-formedness.
Kazanina, Lau, Lieberman, Yoshida, and Phillips
(2007) replicated the gender mismatch effect
using a self-paced reading paradigm and attributed
the effect to the parser’s active search for an antecedent. They suggest that when comprehenders
encounter a cataphoric pronoun in a clause like
While she was working two jobs to pay the bills
they anticipate the upcoming main clause subject
position and construct a referential dependency
between that position and the pronoun. If the
sentence continues with a compatible main
clause subject, such as Kathryn was taking classes
full-time, then comprehension proceeds smoothly.
But if the main clause subject mismatches the
pronoun, as in Russell was taking classes full-time,
then the dependency must be revised, and processing disruption ensues. Kazanina and colleagues
regard this mechanism as a counterpart of the
active dependency formation mechanism that has
been extensively documented for filler-gap dependencies (Frazier & Flores D’Arcais, 1989;
Garnsey, Tanenhaus, & Chapman, 1989; Kaan,
Harris, Gibson, & Holcomb, 2000; Stowe,
1986). We return in the General Discussion to
the issue of the specific mechanisms that underlie
the gender mismatch effect.
Kazanina and colleagues show further that the
gender mismatch effect disappears in syntactic
contexts where backwards anaphora is excluded
by Principle C of the binding theory. Principle C
rules out coreference between a pronoun and any
referring expression that it c-commands
(Chomsky, 1981; Reinhart, 1983).1 They found
that when comprehenders encounter a nominal
in the ccommand domain of a cataphoric
pronoun, as in She was taking classes full-time
while {Kathryn/Russell} was working two jobs to
pay the bills, reading times were unaffected by the
gender compatibility between the nominal and
the pronoun. This finding was confirmed across
several different types of structures that are
subject to Principle C. (For related findings in
Japanese see Aoshima, Yoshida, & Phillips,
2009.) This led Kazanina and colleagues to argue
that Principle C acts as a constraint on generation
in online comprehension (for earlier evidence from
a naming study see Cowart & Cairns, 1987). It is
unlikely that the structurally inaccessible antecedents were ignored in these studies due to insufficient salience, since the critical data came from
reading times when those nominals were in the
focus of attention.
The finding that comprehenders apparently fail
to even consider anaphoric dependencies that
violate Principle C raises the question of why
this constraint should exert such a powerful
effect in online reference resolution, particularly
in light of the inconsistent findings from studies
on other constraints on anaphora. At least two
possibilities suggest themselves. First, there may
be a general advantage for constraints on backwards anaphora, due to the possibility for sequential evaluation of potential antecedents. Second,
the structural properties of Principle C may be
particularly conducive to exclusion of inappropriate antecedents, since the search mechanism
may recognize the irrelevance of an entire
structural domain—that is, any domain that is
c-commanded by the pronoun—in advance of
encountering any bottom-up information about
potential antecedents in the input. We explore
these questions using Russian, which in addition
to Principle C exhibits another Russian-specific
constraint on backwards anaphora. As described
below, the two constraints apply in superficially
similar circumstances yet differ in a number of
important respects, such as in when and how
potential antecedents can be conclusively excluded
from an antecedent search.
1
Apparent exceptions to Principle C have been discussed in the linguistic and psycholinguistic literature—for example, the sentence He was threatening to leave when Billy noticed that the computer had died in which the pronoun he may be understood as coreferential with Billy (Harris & Bates, 2002). Although we believe that these cases may not ultimately be counterexamples to the structural
Principle C (see Kazanina, 2005, for more discussion), we took care to avoid such potential exceptions in constructing the experimental materials for the present study and verified the acceptability of all contrasts under investigation by offline questionnaires.
374
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
Russian, like English, generally allows backwards
anaphora. This is illustrated in Example (1) using
sentences in which an initial subordinate clause is
introduced by the subordinator posle togo kak
“after” (see Examples below). Also like English,
backwards anaphora in Russian is constrained
by Principle C, such that a pronoun cannot
c-command its antecedent. Hence in Example (2)
the pronoun subject of the main clause cannot be
interpreted as coreferential with the subject of the
following subordinate clause. Intended coreference
in the examples is indicated by the use of subscript
indices. Since these two constructions in Russian
parallel their English counterparts, we expect that
manipulation of the gender of the second clause in
each case should elicit the same reading-time
effects as those previously observed in English.
The second, Russian-specific constraint on
backwards anaphora is exemplified by biclausal
sentences such as Example (3) that begin with
the subordinator poka “while” (henceforth, pokasentences). In (3) coreference between the
pronoun on “he” and the name Ivan is disallowed
due to an idiosyncratic constraint on backwards
anaphora in Russian (note that in the English
translation of (3) coreference between he and
Ivan is perfectly acceptable). The constraint,
which we refer to as the poka-constraint for the
sake of exposition, is cross-linguistically rare and
appears not to apply even in other Slavic
languages that are closely related to Russian,
such as Polish or Serbo-Croatian. Critically,
even in Russian the scope of the poka-constraint
is limited: The constraint applies most strongly in
contexts in which both the pronoun subject of the
poka-clause and the following main subject are
agentive, as in (3). Sentences in which one of
the subjects is nonagentive are rather more acceptable on the coreference reading, as exemplified
by Example (4) with an experiencer main
subject (see also Kazanina & Phillips, 2001).
Thus, the Russian poka-constraint can be summarized as follows: In poka-sentences an agentive
pronoun subject of an embedded clause cannot
corefer with an agentive main clause subject.
The constraint also applies in sentences containing other subordinators that encode simultaneity,
such as v to vremja kak “at the time when”.
Note that the poka-constraint also does not
block coreference in sentences in which the
embedded clause expresses a habitual or a
generic event. However, in this paper we do not
consider such cases, and we sought to avoid
them in our experimental materials. More
detailed treatments of the contexts where the
poka-constraint applies and theoretical accounts
of the phenomenon can be found in other
works (Antonyuk & Bailyn, 2008; Avrutin &
Examples (1) – (4):
1.
Posle togo kak oni pročital knigu, Ivani s”el jabloko.
after
he read
book Ivan ate apple
2.
3.
[no-constraint]
“After he read the book, Ivan ate an apple.”
Oni čital knigu, poka Ivani el jabloko.
he
read book while Ivan ate apple
[Principle C]
“He read a book while Ivan ate an apple.”
Poka oni čital knigu, Ivani s”el jabloko.
while he read book Ivan ate apple
[ poka-constraint]
“While he read the book Ivan ate an apple.”
4.
Poka oni čital knigu, Ivani vspomnil,
while he read book Ivan remembered
čto zabyl kupit’ xleba.
that forgot buy bread
“While he read the book Ivan remembered that he forgot to buy some bread.”
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
375
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
Reuland, 2004; Kazanina, 2005; Kazanina &
Phillips, 2001; Reuland & Avrutin, 2005). For
the purposes of the current study it is sufficient to
note that there is agreement that the constraint is
a part of the grammar of Russian, but that it is
not a purely syntactic restriction and that it implicates discourse representations. From the perspective of online language processing the thematic
role restriction on the constraint is particularly
interesting. Since the thematic role of a subject
noun phrase depends upon predicate information,
this restriction implies that the applicability of the
poka-constraint depends on information that may
not be available until after the pronoun – name
sequence. This, together with the nonsyntactic
nature of the constraint, raises the question of
whether the poka-constraint impacts online reference resolution in the same way that Principle C
does, or whether it applies as a late filter on
candidate antecedents. In the studies that follow
we first verify Russian speakers’ awareness of
Principle C and the poka-constraint and investigate their application during online processing
(Experiment 1). We then go on to test whether
the results from the online task were confounded
by plausibility biases (Experiment 2).
The predictions for the online reading study are
straightforward. First, Russian constructions that
have the same coreference possibilities as their
English counterparts should exhibit the same
processing profile as that found in English
studies. This means that in constructions where
cataphora is freely allowed, Russian speakers
should actively construct a cataphoric dependency
upon encountering a potential cataphoric pronoun
in the first clause. This in turn should lead to
processing disruption (mismatch effects) when
the anticipated antecedent turns out to be
incompatible with the pronoun in its gender. In
Russian constructions that are subject to Principle
C readers should avoid predicting an antecedent
in grammatically illicit positions. The behavioural
reflex of this should be a lack of sensitivity to the
gender (mis)match between a pronoun and an
illicit antecedent—that is, no mismatch effect.
Second, predictions for Russian constructions that
are subject to the poka-constraint vary, depending
376
on the way in which the constraint is implemented
online. If all constraints on backwards anaphora are
implemented similarly, as constraints that block the
generation of candidate anaphoric relations, then we
might expect to observe a similar processing profile
to Principle C sentences—that is, lack of mismatch
effects. Alternatively, there are good reasons to
expect that the online impact of the poka-constraint
should be different. Since the constraint blocks coreference between pairs of agentive nominals, but the
thematic role of a subject typically cannot be determined until verb argument structure information is
processed, the constraint may not be triggered until
after a potential antecedent has already been interpreted. This may lead to an online profile where
readers initially construct a cataphoric dependency
that must subsequently be rescinded once verb argument structure confirms that the antecedent is an
agent. If rescinding a cataphoric dependency leads
to processing disruption, then this could lead to
greatest processing difficulty in sentences where
the pronoun and the potential antecedent match
in gender—that is, a gender match effect, the
inverse of the reading time profile predicted for
constructions where coreference is fully acceptable.
This scenario would be particularly interesting, as
it would indicate that two constraints on coreference—Principle C and the poka-constraint—that
apply in superficially similar configurations are
nevertheless implemented differently online. This
in turn would imply that it is not possible to give a
uniform answer to the question of whether constraints on coreference are implemented online as
constraints on the generation of interpretations or
as filters that apply after interpretations have been
constructed.
EXPERIMENT 1
Experiment 1 had two parts: an offline rating task
and an online self-paced reading task. The offline
rating task (Experiment 1A) tested the premise of
our study that Russian speakers should reject coreference between a cataphoric pronoun and a main
clause subject in Principle C or poka-contexts, but
should accept coreference in other backwards
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
anaphora structures that do not violate any
grammatical constraints. The online experiment
(Experiment 1B) investigated the process of establishing backwards anaphoric dependencies in realtime processing. In particular, we were interested
in whether the different types of grammatical
information associated with Principle C and the
poka-constraint cause grammatically inaccessible
antecedents to be removed from consideration in
the same fashion.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Materials and design
The critical experimental items in Experiment 1A
were based on the experimental items from
Experiment 1B: Three of the six conditions from
the online task were used in the offline rating
task. We first describe the materials for the
online experiment (Experiment 1B) and then
explain which conditions were used in the offline
task (Experiment 1A).
The online experiment had six conditions, consisting of four conditions (the Principle C and
poka conditions) organized in a 2 2 factorial
design and two additional no-constraint conditions. A sample set of materials for the online
experiment is shown in Table 1, and a full list of
items is provided in Appendix A. A total of 24
sets of items were constructed according to a
2 2 design with the factors constraint type
(Principle C vs. poka-constraint) and gender congruency (gender match vs. gender mismatch)
affecting the relation between the pronoun in the
first clause and the subject of a second clause.
The biclausal target structures (italicized in
Table 1) were all embedded inside an enclosing
clause, for reasons explained below. The only
difference between the gender-congruent and
gender-incongruent variants in each pair was the
gender of the subject of the second clause (and
the following predicate, due to gender agreement
on Russian past tense verbs). The gender of the
pronoun was counterbalanced across items. The
second subject was always a gender-unambiguous
proper name, and the number of Cyrillic characters and syllables in the gender-matching
and gender-mismatching names was matched.
Materials were designed so as to limit the likelihood of a plausibility bias for or against a coreferential interpretation of the name –pronoun
pair, independent of constraints on anaphora. In
particular, care was taken to use main and
embedded clause events that could plausibly be
performed either by different agents or by a
single agent. Furthermore, pairs of events were
chosen such that the plausibility of the disjoint
or coreference interpretation was similar for the
Principle C and the poka conditions. This was
Table 1. Sample set of items used in Experiment 1
Condition
Russian
English translation
Principle C
conditions, gender
match/gender
mismatch
Poka conditions,
gender match/
gender mismatch
Poskol’ku pered efirom onai prosmatrivala teksty
“Since before the broadcast she looked through the
soobščenij, poka Marina/Daniil grimmirovalas’/
news texts while Marina/Daniel put on makegrimmirovalsja k načalu s”emok, Zojai pervoj uznala
up for the shoot, Zoja was the first one to learn
sensacionnuju novost’.
about the sensational news.”
Poskol’ku pered efirom poka onai prosmatrivala teksty
“Since before the broadcast while she looked
soobščenij, Marina/Daniil grimmirovalas’/
through the news texts Marina/Daniel put on
grimmirovalsja k načalu s”emok, Zojai sama opredelila
make-up for the shoot, Zoja figured out the
porjadok reportažej v vypuske.
order of the reports in the program by herself.”
No constraint, gender Xotja posle togo kak onai napisala zakazannuju stat’ju,
“Although after she wrote the commissioned article
match/gender
Natašai/Mixail pravila/pravil tekst neskol’ko raz,
Natasha/Michael edited the text several times,
mismatch
Michael/Natasha was most proud of the
Mixail/Natašai bol’še vsego gordilsja svoim
pervonačal’nym variantom.
original version.”
Note: The critical name in the second subject position is shown in bold. The biclausal target structures are shown in italics. Subscript
indices indicate the intended licit backward anaphoric dependencies. The acceptability rating task (Experiment 1A) used only
sentences in which the second subject matched the gender of the pronoun.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
377
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
important in order to exclude the possibility that
the reading-time profiles in the two pairs of conditions might differ due to a bias in the stimuli
rather than to differences in the online application
of the two constraints. Experiment 2 further
addresses this issue.
Additionally, the experiment contained 12
sentences from the no-constraint conditions. The
no-constraint conditions were licit backwards
anaphora sentences such as (1) that contained an
embedded clause introduced by the subordinators
do togo kak “before” or posle togo kak “after”.
These embedded clauses linearly preceded the
main clause and were therefore structurally parallel
to the poka conditions. Similar to the other two
conditions, the gender of the second subject NP
was varied to obtain gender-congruent and
gender-incongruent variants. As in the Principle
C and poka-constraint conditions the gender of
the pronoun was counterbalanced across items.
These structures do not violate any constraints
on coreference and were included to provide a
baseline measure of how Russian speakers establish licit backwards anaphoric dependencies. In
light of the previous findings in English we
expected a gender mismatch effect at the second
subject position.
The lexical content of the no-constraint
conditions differed from the Principle C and
poka conditions, because of the differing semantic
conditions on the use of the subordinators used.
The no-constraint conditions used the subordinators “before” and “after” and thus described events
that occurred in succession. The Principle C
conditions and the poka-constraint conditions
used the subordinator “while” and thus described
events that overlapped in time. Consequently, a
basic semantic feature of the relation between
the events in the sentences had to differ in the
no-constraint conditions. Similarly, the subordinators “before” and “after” in Russian require
different verbal aspectual forms (perfective) than
the configurations that trigger the “poka” constraint in sentences with “while” (imperfective),
and hence lexical matching of the verbs was not
possible. It was considered more important to
maximally control the lexical material and
378
plausibility between the two pairs of constraint
conditions, since the primary goal of the study
was to compare the impact on online interpretation of the two constraints on cataphora.
In order to ensure that the cataphoric pronoun
could be associated with an acceptable antecedent
in every sentence, all target structures were
embedded in a further clause introduced by the
subordinators xotja, “although”, or poskol’ku,
“since”. The gender of the third-clause subject
was chosen such that each sentence had a unique
acceptable antecedent for the pronoun. This
follows a design strategy that proved successful
in previous studies on English (Kazanina et al.,
2007). The lexical content of the third clause was
varied between the Principle C and the poka conditions within the same set, in order to maximize
the overall plausibility of the sentences. In the
Principle C and poka conditions the subject of
the third clause matched in gender with the
pronoun and served as an acceptable antecedent.
In the no-constraint conditions the gender of the
third-clause subject mismatched the pronoun in
the gender-match condition, due to the availability
of coreference between the pronoun and the
second-clause subject, but the third-clause
subject matched the gender of the pronoun in
the gender-mismatch condition.
EXPERIMENT 1A: ACCEPTABILITY
RATING TASK (OFFLINE)
The offline rating task tested the prediction that
Russian speakers should reject coreference between
a cataphoric pronoun and a main clause subject in
Principle C or poka-constraint contexts, but
should accept coreference in other backwards anaphora structures that do not violate any constraints.
Method
A total of 33 native speakers of Russian from
Moscow participated in the experiment.
Participants saw three-clause sentences in which
the pronoun and the second subject were highlighted in bold face and were asked to judge the
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
acceptability of a coreferential interpretation of the
boldfaced pronoun–name pair, using a scale from 1
(impossible) to 5 (absolutely natural). The specific
form of the question (translated from Russian)
was: Can the pronoun in bold and the noun in bold
refer to the same person? Only gender-matching variants of each condition were presented. In addition,
in the no-constraint condition the gender of the
third subject was changed to match the gender of
the pronoun. As a result, in all three conditions
the pronoun was followed by two gender-matching
nouns, thereby ensuring that any differences across
conditions in the rating of the critical second subject
would not be confounded by the number of other
gender-matching antecedents in the sentence.
The experiment was administered in the form
of a pen-and-paper questionnaire using a similar
methodology to Gordon and Hendrick (1997). A
total of 24 sets of Principle C and poka-constraint
conditions were distributed among two lists
according to a Latin Square design; each list
additionally contained the 12 no-constraint sentences. Each participant saw only a third of all
experimental items from one of the lists—that is,
4 sentences from each of three experimental
conditions (Principle C, poka-constraint, noconstraint) interspersed with 4 control and 4
filler sentences. The controls were sentences such
as (5) with licit backwards anaphora. The filler
sentences were structures with licit forwards
anaphora such as (6). In total each questionnaire
contained 12 experimental items (four per
condition) interspersed with 8 fillers.
5.
6.
Backwards anaphora (BA) control
V naznačennyj den’ Astaxov, kotoryj
obeščal emu pomoč’ v ustanovke novogo
programmnogo obespečenija, tak i ne
pojavilsja v novom ofisnom centre Andreja
Isaeva.
“On the due date Astakhov, who had
promised to help him in installing the software still didn’t show up at Andrej Isaev’s
new office.
Forwards anaphora (FA) filler
Nesmotrja na to, čto Irina nikogda ne
sobiralas’ stat’ balerinoj, vysokaja ocenka
Ekateriny pol’stila ej i byla ochen’ prijatnoj.
“Although Irina was never going to
become a ballerina, the high praise from
Ekaterina flattered her and was very
pleasant.”
Results and discussion
Fillers with forwards anaphora received an average
rating of 4.6/5. Mean rating scores for experimental conditions and the control are given in Table 2.
Mean ratings were very low in the Principle C
condition and the poka condition and much
higher in no-constraint and the backwards anaphora control conditions. A linear mixed-effects
model with participants and items as random
factors and condition as a fixed factor revealed a
strong effect of condition, as witnessed by the
fact that the model with the factor condition was
superior to a simpler model that did not include
that factor: x2(3) ¼ 76.1, p , .001. Planned pairwise comparisons showed that the scores in the
no-constraint condition did not differ significantly
from those in the backwards anaphora control,
t(262) ¼ –1.1, p ¼ .257, whereas it was rated significantly higher than the Principle C condition,
t(262) ¼ –13.0, p , .001, and the poka condition,
t(262) ¼ –8.8, p , .001. The difference in ratings
between the Principle C and the poka condition
was significant, t(262) ¼ –2.8, p ¼ .005, despite
the fact that both conditions received consistently
low ratings. The slightly higher ratings for the
poka-constraint sentences may reflect their
structural similarity to the far more acceptable sentences in the no-constraint conditions. (However,
the small acceptability difference should not itself
Table 2. Mean acceptability rating scores by condition in
Experiment 1A
Score
Condition
M
SE
Principle C
Poka-constraint
No constraint
BA control
1.5
1.9
3.6
3.9
0.10
0.11
0.12
0.11
Note: BA ¼ backwards anaphora.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
379
KAZANINA AND PHILLIPS
lead to an expectation for strongly divergent online
effects of the two constraints.)
In sum, the acceptability ratings confirm the
prediction that Russian speakers should disallow
coreference in sentences that are subject to
Principle C and the poka-constraint, whereas
they accept backwards anaphora in the no-constraint sentences.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
EXPERIMENT 1B: SELF-PACED
READING EXPERIMENT
Method
Participants
A new group of 48 Russian speakers (ages 18– 28
years) was recruited in Moscow. All participants
had normal or corrected-to-normal vision and no
history of language disorders. They gave informed
consent and were paid the equivalent of $10 for
their participation.
Procedure
A total of 24 sets of items consisting of a gendermatch and a gender-mismatch variant of the
Principle C conditions and of the poka-constraint
condition (i.e., four conditions per set) were
distributed among four presentation lists in a
Latin square design. A total of 12 items from
the no-constraint conditions were added to each
presentation list so that each list contained one
variant of each item (yielding six gender-match
and six gender-mismatch variants per list). Each
list also contained 84 filler sentences that varied
in length and complexity and were superficially
similar to the target items—for example, they
contained proper names or started with a subordinator. Participants were randomly assigned
to one of the lists, and the order of the stimuli
within a presentation list was randomized for
each participant.
Participants were tested using a Windows
laptop computer running the Linger software
(developed by Doug Rohde, MIT). Sentences
were presented in a standard noncumulative
word-by-word moving window paradigm (Just,
380
Carpenter, & Woolley, 1982) using the font
Courier New Cyrillic 20. Each sentence initially
appeared on the screen masked by dashes that
covered all letters and punctuation marks, but
left spaces between words unmasked. Each time
the participant pressed the spacebar a new word
appeared on the screen, and the previous word
was remasked by dashes. Care was taken to
ensure that all words from the critical second
clause appeared on the same line of the display
(the first line). A yes/no comprehension question
appeared after each sentence in a single display.
Comprehension questions were designed so that
they were equally likely to question different
parts of the sentence (i.e., first, second, or third
clause). In order to conceal the main manipulation
of the study and reduce the risk of conscious
reading strategies the comprehension questions
never targeted the pronoun interpretation.
Participants were instructed to read sentences at
a natural pace and to respond to the comprehension questions as accurately as possible. If the
question was answered incorrectly the word
“Incorrect” briefly appeared in the centre of the
screen. The testing session lasted approximately
40 minutes.
Data treatment
In order to ensure sufficient numbers of data points
in each cell of the analysis, participants were
excluded from the final analyses if they failed to
correctly answer at least half of the comprehension
questions in any individual experimental condition, even if they had a high overall accuracy
rate. This led to the exclusion of two participants.
The final analyses were based on the data from 46
participants. Only sentences for which the comprehension question was answered correctly were
included in the analyses of reading times.
Furthermore, reading times that exceeded a
threshold of 2,500 ms were replaced by the
threshold value. This winsorizing procedure
affected 0.4% of data points in the Principle C
conditions and in the poka-constraint conditions
and 0.3% of data points in the no-constraint
conditions.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
Reading times were analysed in regions that
corresponded to a single word, with the exception
of the last region in each clause, which represented
the average reading time per word for all remaining words in that clause. Reading times were statistically analysed by fitting a linear mixed effect
model using the lmer function from the lme4
package in R (Version 2.6.2; CRAN project; R
Development Core Team, 2009. Unlike more traditional analyses of variance (ANOVAs), mixed
effects models take unaveraged data as input and
make it possible to incorporate random effects of
both participants and items within a single analysis
(for more information on the use of mixed effects
models in psycholinguistics see Baayen, 2008).
Models were fitted using a restricted maximum
likelihood technique. For the data from the
poka-constraint and Principle C conditions the
model fitting proceeded as follows: Initially a
model that only included the random factors (participants and items) was applied. This initial
model was next enriched by adding the first fixed
factor constraint (Principle C vs. poka-constraint)
and subsequently by including the other fixed
factor gender congruency (gender match vs.
gender mismatch). Finally the interaction
Constraint Gender Congruency was added to
the model. Each successive pair of models was
evaluated to assess whether the additional factor
improved the model fit to the data. The most
complex model that significantly improved the fit
over the previous model is considered to be the
best fitting model, and its estimates are reported
below. In all cases where an interaction was significant we report pairwise comparisons and the relevant 95% confidence interval derived by Markov
Chain Monte Carlo simulation (Baayen,
Davidson, & Bates, 2008). The model-fitting procedure was identical for the no-constraint conditions, with the exception that only one fixed
effect (gender congruency) was included. Model
fitting was performed for each region in the sentence individually. For the analyses of the binary
variable “accuracy (correct/incorrect)” a binomial
family was used.
Since exclusion of 2 participants led to an unbalanced number of participants across item lists an
additional set of models were tested that included
item list as an additional factor. These analyses
showed no impact of the list factor, and hence
that factor is not considered further. All significant
main effects and interactions with p , .05 are
reported.
Results
Comprehension question accuracy
The mean comprehension question response accuracy was 88.6% in experimental items and 93.2% in
filler items. Accuracy rates for individual experimental conditions are given in Table 3.
In the no-constraint conditions there was a
marginally significant effect of gender congruency
(z ¼ 1.904, p ¼ .057) due to a higher mean accuracy rate in the gender-match condition than in
the gender-mismatch condition. The effect of
gender congruency was also significant in the
Principle C and poka-constraint conditions
(z ¼ – 2.9, p ¼ .003) but the direction of the
effect was reversed: A higher mean accuracy rate
was found in the gender-mismatch condition
than in the gender-match condition (91.3% vs.
86.2%). There also was a marginally significant
effect of constraint type (z ¼ 1.8, p ¼ .068) due
to higher accuracy rates in the Principle C conditions than in the poka-constraint conditions
(90.2% vs. 87.3%).
Table 3. Mean comprehension question accuracy rates by condition
from Experiment 1B
Accuracy
Condition
No constraint
Gender match
Gender mismatch
Principle C
Gender match
Gender mismatch
Poka
Gender match
Gender mismatch
M
SE
92.8
88.0
1.6
2.0
88.0
92.4
2.0
1.6
84.4
90.2
2.2
1.8
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
381
KAZANINA AND PHILLIPS
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Reading times (RTs)
The results from the no-constraint conditions are
reported first, followed by the results from the
Principle C and poka conditions. The region
means and information on the best fitting model
for each region are summarized in Appendix B.
No-constraint conditions
Figure 1 shows word-by-word reading times in the
no-constraint conditions. Mean reading-time
differences were small throughout the first clause
(Regions 1 – 8), at the critical second subject
(Region 9: Natasha/Michael) and in the following
region. However, in Region 11, two words after
the critical second subject NP, the mean reading
time in the gender-incongruent sentences was
65 ms longer than that in the gender-congruent
sentences (gender congruent, 542 ms; gender
incongruent, 607 ms). This difference led to a marginally significant main effect of gender congruency
(95% confidence interval, CI ¼ +50 ms), b ¼
–48.3, t(497) ¼ –1.86, p ¼ .057. The same
numerical trend for longer reading times in the
gender-incongruent condition continued at the
following two regions (Regions 12 and 13), but
reached significance only in Region 13 (gender
congruent, 542 ms; gender incongruent, 614 ms;
mean difference, – 72 ms; 95% CI ¼ +55 ms),
b ¼ – 57.6, t(497) ¼ – 2.06, p ¼ .040. There were
no other significant effects in the remainder of the
sentence. Thus, reading times in the no-constraint
conditions replicate the gender-mismatch effect
observed in studies of cataphora processing in
English (Kazanina et al., 2007; van Gompel &
Liversedge, 2003).
Principle C and poka-constraint conditions
Figure 2 shows the results from the Principle C
conditions, and Figure 3 shows the results from
the poka-constraint conditions.
In Region 2 there was an unexpected significant
effect of constraint type, b ¼ 30.3, t(976) ¼ 2.1,
p ¼ .030,
and
a
marginally
significant
Constraint Gender Congruency interaction,
b ¼ – 36.3, t(976) ¼ –1.8, p ¼ .069. This effect
must be spurious, since the conditions were identical up to this point, and it should be noted that
the effect corresponded to a small difference in
mean reading times. At Region 6 of the first
clause there was also a marginally significant
effect of constraint type, b ¼ 57.5, t(976) ¼ 1.94,
p ¼ .057, which was most likely due to the
Figure 1. Mean reading times (in ms) from the no-constraint conditions in Experiment 1B. A gender mismatch effect (GMME) that
appeared two words after the critical subject noun phrase (NP) is marked by a box.
382
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
Figure 2. Mean reading times (in ms) from the Principle C conditions in Experiment 1B.
presence of an extra word in the poka-constraint
conditions compared to the Principle C conditions. There were no other significant effects or
interactions at any region in the first clause.
There was a main effect of constraint type at the
critical second subject in Region 9 due to longer
mean reading times in the poka conditions than in
the Principle C conditions (mean reading times
555 ms vs. 499 ms, respectively), b ¼ –59.4,
t(978) ¼ –4.04, p , .001. This effect was likely
caused by differences in the lexical material in the
immediately preceding region. The subordinator
while immediately preceded this region in the
Principle C conditions and may have provided an
effective cue for the appearance of a subject NP in
Region 9. There were no significant effects or interactions in Regions 10–11. However, in Region 12,
three words after the critical second subject, there
was a significant effect of gender congruency,
b ¼ 57.8, t(960) ¼ 2.48, p ¼ .015, and a marginally
significant Constraint Gender Congruency interaction, b ¼ –61.7, t(960) ¼ –1.89, p ¼ .057.
Pairwise comparisons within each level of the constraint factor revealed a significant effect of congruency in Region 12 of the poka conditions due
to longer mean reading times when the second
subject matched in gender with the preceding
pronoun (gender congruent, 581 ms; gender
incongruent, 523 ms; mean difference ¼ 58 ms;
95% CI ¼ +48 ms), b ¼ 58.1, t(473) ¼ –2.38,
p ¼ .018. This is a gender match effect, the inverse
of the effect found in the no-constraint conditions.
In the Principle C conditions, on the other hand,
reading times were almost identical at both levels
of the gender congruency factor (gender congruent,
528 ms; gender incongruent, 533 ms; mean
difference ¼ –5 ms; 95% CI ¼ +42 ms), b ¼
–2.0, t(487) ¼ –0.09, p ¼ .957. There were no
significant effects in Region 13.
At the subject of the third clause in Region 14
there was a main effect of gender congruency,
b ¼ 45.9, t(977) ¼ 2.85, p ¼ .006. In both the
Principle C conditions and the poka-constraint
conditions mean reading times for the third
subject NP were slower in conditions where the
earlier second subject NP had matched the
gender of the pronoun. Recall that in all four of
these conditions the third subject NP provided a
grammatically accessible antecedent for the
pronoun. In Region 15 there was a main effect of
congruency, b ¼ 52.9, t(976) ¼ 2.76, p ¼ .006, and
a significant Constraint Gender Congruency
interaction, b ¼ –74.3, t(976) ¼ –2.76, p ¼ .005.
Resolution of this interaction revealed that the
effect of congruency was significant in the poka
conditions, due to longer mean reading times in the
gender-congruent condition than in the genderincongruent condition (527 vs. 473 ms, respectively;
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
383
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
Figure 3. Mean reading times (in ms) from the poka conditions in Experiment 1B. A gender match effect (GME) that started three words
after the critical subject noun phrase (NP) is marked by a box.
mean difference ¼ 54 ms; 95% CI ¼ + 39 ms),
b ¼ 51.8, t(480) ¼ 2.59, p ¼ .013, but not in
the Principle C conditions (gender congruent
– 477 ms; gender incongruent – 502 ms; mean
difference ¼ – 25 ms; 95% CI ¼ + 33 ms), b ¼
–14.7, t(496) ¼ – 0.86, p ¼ .397. There were no
significant or marginally significant main effects
or interactions in the remainder of the sentence
(Regions 16 – 17).
Discussion
In Experiment 1B we found that manipulation of
the gender of the subject of the second clause led
to a gender mismatch effect in the no-constraint
conditions and had no impact on reading times
in the Principle C conditions. These results from
Russian replicate earlier findings based on similar
English sentences (Kazanina et al., 2007; van
Gompel & Liversedge, 2003). Additionally, we
found a gender match effect at the same region
in the poka-constraint conditions, which are subject
to a Russian-specific grammatical constraint. This
gender match effect is the inverse of the readingtime pattern observed in the no-constraint conditions.
It should be noted that the comprehension
accuracy results were largely consistent with the
reading-time results. In the no-constraint
384
conditions the gender-matching condition
yielded a higher accuracy rate than its gender-mismatching counterpart, whereas the pattern was
reversed in the other two pairs of conditions.
This is likely because in the no-constraint condition the gender-matching condition was the
easier sentence to process: The parser actively
anticipated an antecedent in the second subject
position, and this expectation was fulfilled. In
the Principle C and the poka conditions, on the
other hand, the second subject was easier to rule
out in the gender-incongruent condition when it
was “doubly” illicit, due to the constraint and to
the mismatch in gender with the pronoun.
It is interesting that the Principle C and the
poka-constraint conditions yielded different
reading-time profiles, despite the fact that both
structures are quite similar on the surface and
yielded similarly low acceptability ratings for backwards anaphora interpretations. Before interpreting
these results, however, it is important to ensure that
the differences in the processing of the two pairs of
conditions were indeed due to the different constraints on coreference, rather than to artefacts of
the stimuli that we used. The sentences in the
Principle C and poka conditions differed only in
the position of the subordinator while. Ideally,
this manipulation should have had no other effect
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
beyond changing the sentence structure so that the
sentence invoked either Principle C or the pokaconstraint. However, this structural change was
accompanied by a shift in the figure/ground
relation of the two events (E1 and E2) described
in the sentence, as illustrated in (7):
7.
Principle C:
(10) and (11) illustrate an event pair that shows a
figure/ground symmetry when the events have
different agents, but which is less clearly symmetric
when the events are performed by the same agent.
10.
a. Jane was feeling dizzy while Bill was
cleaning the floor.
[E1 figure, E2 ground]
E1, while E2
[E1 ¼ figure, E2 ¼ ground]
Downloaded by [University Of Maryland] at 05:32 21 October 2011
poka condition: While E1, E2
[E1 ¼ ground, E2 ¼ figure]
The subordinator while establishes a relation in
which the embedded event serves as a ground for
the main event, which is a figure (also known as foregrounding/backgrounding, e.g., Matthiessen &
Thompson, 1988). Thus, in addition to the structural differences between the conditions in (7), the
sentences also differed in which of the events E1
and E2 served as figure or ground. This change
of figure/ground relations could be important,
because many pairs of events do not have the property that either event can naturally serve as figure
while the other takes the role of ground in a whileclause. A pair of events that meet this criterion is
illustrated in (8), whereas the events in (9) do not.
8.
E1 ¼ reading a letter, E2 ¼ eating an apple
a. Jane read a letter while Bill ate an apple.
[E1 figure, E2 ground]
b. While Jane read a letter, Bill ate an apple.
[E1 ground, E2 figure]
9.
E1 ¼ breaking a glass, E2 ¼ eating an apple
a. Jane broke a glass while Bill ate an apple.
[E1 figure, E2 ground]
b. While Jane broke a glass, Bill ate an apple.
[E1 ground, E2 figure]
All stimuli used in Experiment 1B were constructed using symmetric event pairs like the pair
in (8). However, it is important to also ensure
that the pair of events preserves its figure/ground
symmetry when the two events are performed by
the same agent. This is important, since a semantic
bias against a coreferential interpretation of the two
subject NPs might lead to spurious evidence for the
effects of a linguistic constraint. The examples in
E1 ¼ feeling dizzy, E2 ¼ cleaning the floor
b. While Jane was feeling dizzy, Bill was
cleaning the floor.
[E1 ground, E2 figure]
11.
E1 ¼ feeling dizzy, E2 ¼ cleaning the floor
a. Janei was feeling dizzy while shei was
cleaning the floor.
[E1 figure, E2 ground]
b. While Janei was feeling dizzy, shei was
cleaning the floor.
[E1 ground, E2 figure]
Example 10 shows that the pair of events E1 and
E2 is symmetric. When events E1 and E2 are performed by different agents, either of them can
serve as a semantically plausible background for
the other. Hence, (10a) and (10b) are equally
plausible. However, the symmetry disappears if
the events are performed by the same person, as
in (11). (11a) is quite natural on the coreference
reading, and it suggests that E1 and E2 can, in
principle, be performed simultaneously by the
same agent. However, the coreference reading in
(11b) is less plausible, for the reason that feeling
dizzy is a not an ideal setting for an event of cleaning the floor in a setting where these events describe
a simultaneous activity by the same person.
The existence of contrasts like the one in (10)–
(11) raises the question of whether the materials in
Experiment 1 might have contained similar biases.
We therefore sought to test whether the pairs of
events described in each set were such that switching
their figure/ground relation would not introduce a
semantic bias towards a coreference or disjoint
interpretation, as in (11). If both figure/ground
combinations of the events were equally plausible
under a coreference interpretation, we could more
confidently attribute the processing differences
between the Principle C and the poka conditions
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
385
KAZANINA AND PHILLIPS
to the fact that they invoke different constraints on
coreference. Experiment 2 addresses this issue.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
EXPERIMENT 2
Experiment 2 was a pencil-and-paper questionnaire. Its aim was to test whether the pairs of
events used in the materials for the Principle C
and poka-constraint conditions in Experiment 1
were balanced and whether inverting their figure/
ground configuration would create a bias for a coreferential or disjoint interpretation of the two
subject NPs. If there were independent reasons to
favour or disfavour a coreferential interpretation
of the test sentences, then it would be inappropriate
to attribute the reading-time profiles observed in
Experiment 1B to effects of grammatical constraints. Additionally, we wanted to ensure that
all sentences used in Experiment 1 were amenable
to a coreference interpretation in all respects
except for the effect of constraints on coreference.
Method
Participants
Experiment 2 was administered to 24 native speakers of Russian residing in Moscow (10 males, 14
females; mean age 30.8 years, range 18– 45 years).
Materials and design
All gender-matching variants of the Principle C
and poka-constraint conditions from Experiment
1 were used in Experiment 2, yielding 24 sets of
two conditions. The sentence-initial subordinators
although or since and the enclosing third clause were
discarded. Furthermore, the order of the pronoun
and the name was reversed to obtain forwards anaphora, as illustrated in Table 4. Consequently,
stimuli from the Principle C conditions were transformed into forwards anaphora sentences in which
the main clause preceded the embedded clause
(“forwards anaphora, main first”), whereas stimuli
from the poka conditions were transformed into
sentences with forwards anaphora in which the
embedded clause preceded the main clause (“forwards anaphora, embedded first”). Changing the
pronoun – name order used in Experiment 1 to forwards anaphora sentences with a name –pronoun
order rendered Principle C and the poka-constraint
irrelevant and thus made it possible to identify any
potential effect of the figure/ground relation of the
events on sentence interpretation.
Participants rated the plausibility of each sentence using a scale from 1 (impossible) to 5 (absolutely natural). The critical name and pronoun were
highlighted in bold, and participants were explicitly
told that they referred to the same person. A total of
24 pairs of target items were distributed among two
lists using a Latin square design and were interspersed with 12 control sentences, half of which
were highly plausible (“control, highly plausible”)
or highly implausible (“control, highly implausible”) on the coreference reading. These controls
were designed to mask the target sentences, and
they served as a measure that the participants
understood the task correctly.
Table 4. Sample set of items from Experiment 2
Russian
Marina prosmatrivala teksty soobščenij, poka
ona grimmirovalas’ k načalu s”emok.
Poka Marina prosmatrivala teksty soobščenij,
ona grimmirovalas’ k načalu s”emok.
Tak kak za poslednie tri goda ona ni razu ne brala
otpuska, Olesja tverdo rešila, čto v etom godu
uedet otdyxat’ na more.
Control, highly implausible Poka Inna naxodilas’ po bedro v gipse, ona bez
truda begala po lestnicam.
Forwards anaphora,
main first
Forwards anaphora,
embedded first
Control, highly plausible
English translation
“Marina looked through the news texts while she put
on make up for the shoot.”
“While Marina looked through the news texts she
put on make up for the shoot.”
“Since in the last three years she had never taken
time off, Olesja firmly resolved to go to a seaside
resort.”
“While Inna’s leg was in a cast, she could easily climb
stairs.”
Note: The participants were explicitly instructed to consider the name and the pronoun in bold as referring to the same person.
386
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
Table 5. Mean plausibility ratings in Experiment 2
Condition
M
SE
Forwards anaphora, main first
Forwards anaphora, embedded first
Control, highly plausible
Control, highly implausible
3.2
3.1
4.5
1.3
0.15
0.15
0.10
0.10
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Results and discussion
The results of the rating test are summarized in
Table 5.
As expected, the plausible control sentences
received the highest coreference score, and the
implausible controls received the lowest score.
The forwards-anaphora conditions yielded mean
ratings in the middle of the scoring range,
suggesting intermediate plausibility of these sentences. Linear mixed-effects modelling revealed a
significant effect of condition as a whole, F(3,
860) ¼ 24, p , .0001. Importantly, a planned
comparison of ratings in the two forwards anaphora conditions shows that the ratings for the
main-first and the embedded-first forwards anaphora conditions did not significantly differ,
t(574) ¼ 1.45, p ¼ .149. Experiment 2 therefore
confirmed that the stimuli used in the Principle
C and poka conditions were balanced in terms of
the figure/ground relation between the main and
embedded events. There was no bias towards
higher or lower plausibility of the coreference
reading associated with reversal of the figure/
ground relation of the events in the two
conditions. We can therefore be more confident
that the different reading-time profiles observed
in Experiment 1B reflect differences in the
constraints that apply to each condition.
GENERAL DISCUSSION
We investigated the processing of backwards anaphora in Russian with the goal of understanding
whether grammatical constraints impact the
online search for antecedents for pronouns in a
uniform manner. A self-paced reading experiment
(Experiment 1B) investigated the processing of
three types of structures containing a cataphoric
pronoun in the first clause. In the no-constraint
conditions the subject of the second clause was a
licit antecedent for the pronoun. In the remaining
conditions the second subject was an illicit antecedent due to a constraint on coreference, either
Principle C or the Russian-specific pokaconstraint. Following a strategy used in previous
studies of cataphora processing in English, the
gender of the NP in the second subject position
was manipulated such that it either matched or
mismatched the gender of the cataphoric
pronoun. The logic of this design is that in any
condition where a cataphoric dependency is
attempted we should observe an effect of gender
congruency at or shortly after the critical NP.
The parser’s behaviour in the regions following
the critical second subject was different in each
of the three condition pairs. In the no-constraint
conditions the reading times following the
second subject NP were longer when that NP mismatched in gender with the anaphoric pronoun (a
gender mismatch effect). There was no effect of
gender congruency on the second subject in the
Principle C conditions. Finally, in the Russianspecific poka-constraint conditions reading times
were longer when the second subject matched in
gender with the preceding pronoun (gender
match effect). Experiment 2 confirmed that these
results are unlikely to be due to a plausibility bias
in the pairs of events chosen for Principle C and
poka-constraint conditions.
The gender mismatch effect found in the noconstraint condition in Russian replicates earlier
findings by van Gompel and Liversedge (2003)
and by Kazanina et al. (2007) for similar cases
in English, and it suggests that across languages
the parser uses a similar mechanism to search
for an antecedent for a cataphoric pronoun.
The gender-matching no-constraint conditions
feature a dependency between a pronoun and a linearly following antecedent. In the absence of a previously mentioned discourse referent the pronoun
is temporarily left without a referent, and the
parser initiates a search for a licit antecedent in
the upcoming input. The gender mismatch effect
in the no-constraint conditions reflects the
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
387
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
parser’s surprise upon encountering a genderincongruent nominal in a position where an antecedent was expected. Importantly, as van Gompel
and Liversedge point out, the existence of the
gender mismatch effect implies that the parser is
not hesitant in forming cataphoric dependencies.
In order to be disrupted upon recognizing that a
noun phrase in the input is gender incongruent,
the parser must have initiated dependency formation prior to that moment.
This observation about the earliness of cataphoric dependency formation could reflect either
of two mechanisms. Van Gompel and Liversedge
(2003) suggest that it is a consequence of an architectural property of the parser that makes word
category information about incoming words available before more fine-grained semantic information. In contrast, Kazanina and colleagues
regard early formation of cataphoric dependencies
as a counterpart of “active” dependency formation
found in the processing of filler-gap dependencies
(Crain & Fodor, 1985; Frazier & Flores d’Arcais,
1989; Garnsey et al., 1989; Stowe, 1986; Traxler &
Pickering, 1996). They propose that dependency
formation may be initiated as soon as a cataphoric
pronoun is encountered and before a candidate
antecedent appears in the input. A mechanism of
this kind presupposes the possibility of predictive
structure building, but it also avoids the need to
assume that an architectural constraint makes syntactic category information available more quickly
than other properties of incoming words. Both of
these accounts of the gender mismatch effect are
compatible with the findings presented here.
Moreover, we contend that the two mechanisms
are in fact rather similar, since the mechanism
proposed by van Gompel and Liversedge must
be active to some degree. Even if cataphoric
dependency formation is attempted only after a
candidate antecedent is encountered in the
bottom-up input, the parser must already be in a
control state that encodes the fact that it is searching for an antecedent for a pronoun. Without such
a mechanism, cataphoric dependency formation
could only arise from a massively redundant mechanism that would search backwards for a pronoun
every time that it encounters a noun phrase, an
388
unlikely state of affairs. In addition the control
state would need to be sensitive to which structural
positions the parser does and does not treat as
suitable antecedent positions. This sensitivity is
needed to capture the sensitivity of the gender
mismatch effect to grammatical constraints on
cataphora, as found in this and other studies.
The gender mismatch effect found in the noconstraint conditions also confirms that cataphoric
dependencies are acceptable in Russian. If there
were a general prohibition against cataphoric
dependencies in Russian then we should not
have expected to find that Russian speakers
actively attempt to construct such dependencies
in online processing. Although the gender mismatch effect did not occur immediately upon presentation of the second subject noun, the observed
two-word delay is compatible with results from
previous studies in English. Van Gompel and
Liversedge (2003) found a one-word delay for
the mismatch effect in their eye-tracking studies,
and our own earlier self-paced reading studies in
English showed either an immediate mismatch
effect or a one-word delay. Badecker and
Straub’s self-paced reading studies on forwards
anaphora showed a one – two-word delay in the
effects of gender congruency (Badecker &
Straub, 2002).
Turning to the Principle C conditions, the lack
of any effect of the gender manipulation in the
second subject in these conditions replicates
earlier findings in English (Kazanina et al.,
2007). In both languages we take the lack of an
effect as evidence that the parser does not consider
the second subject as a potential antecedent position for the cataphoric pronoun. This does not
mean that the parser fails to actively search for
an antecedent in these conditions, only that it
does not expect an antecedent in positions that
are restricted by Principle C. This apparent blindness to the possibility of coreference in the
Principle C conditions cannot easily be dismissed
as a consequence of the discourse salience of the
second subject NP, since the crucial reading-time
measures were taken at a point in time when
that NP was in the focus of attention. The
Russian findings add cross-linguistic evidence to
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
Downloaded by [University Of Maryland] at 05:32 21 October 2011
PROCESSING RUSSIAN CATAPHORA
earlier conclusions from English and lend further
support to the view that Principle C is a crosslinguistically robust constraint on structure generation. In other words, rather than generating
cataphoric dependencies that violate Principle C
and subsequently rejecting them based on application of a filter, such illicit coreference patterns
are never even considered by the parser. More generally, early sensitivity to Principle C can be added
to a growing body of evidence that grammatically
accurate representations can be built in real time
(Crain & Fodor, 1985; Stowe, 1986; Traxler &
Pickering, 1996, among others).
Replication of the English findings in the noconstraint conditions and the Principle C conditions provided the background for the most
interesting new contribution of the current study.
The poka-constraint conditions offered a comparison that has not been possible in previous studies
in other languages, as it was possible to compare
the reading-time profiles of sentences that are
impacted by two different constraints on backwards anaphora. This parallel makes it possible
to test whether constraints on coreference apply
in a uniform fashion. The Principle C conditions
and the poka-constraint conditions had many
features in common: The dependent elements
were the same pronouns (on, “he”, or ona, “she”);
the lexical material was identical; the thematic
roles of all nominals were the same, and plausibility was matched across conditions; the linear
distance between the pronoun and the potential
antecedent was almost identical. In addition, the
rating study (Experiment 1A) showed that the
two constraints led to similarly low acceptability
scores. Furthermore, the two constraints were
tested using a within-subjects design. Despite
these parallels, the reading-time profiles were
quite different, with a gender match effect in the
poka-constraint conditions and no effect of
gender congruency in the Principle C conditions.
This implies that the two constraints impact
online interpretation in different ways.
A number of previous studies have investigated
whether constraints on anaphora impact language
processing by blocking the generation of illicit
interpretations or by filtering out illicit
interpretations after they have been constructed.
Most previous studies in this domain have
focused on forwards anaphora. Nicol and
Swinney argued that constraints on reflexives and
pronouns have parallel early impacts on antecedent
activation, potentially supporting an interpretation
of the constraints as constraints on the generation
of interpretations (Nicol & Swinney, 1989). In
contrast, Badecker and Straub argued that the
same constraints act as later filters on candidate
interpretations, although the results from their
series of studies were more equivocal (Badecker
& Straub, 2002). Runner and colleagues used
eye-fixation patterns to argue for differences in
the processing of reflexives and pronouns
(Runner et al., 2006), but their study focused on
special syntactic contexts (“picture noun phrases”,
such as Harry’s picture of himself ) and ultimately
concluded that reflexives in picture noun phrases
are exempt from standard grammatical constraints
on reflexives. Thus, the current study shows more
unequivocally than previous studies that constraints on coreference do not impact language
processing in a uniform fashion.
The gender match effect following the critical
second subject in the poka conditions implies that
participants at least fleetingly considered that
subject as a potential antecedent for the preceding
pronoun. We interpret the slowdown in the
gender-congruent condition as reflecting difficulty
that arises when the parser is forced to retract its
initial consideration of an analysis in which the
name is the antecedent of the preceding pronoun.
If the parser were simply blind to the possibility
of coreference between the pronoun and the
name, then we should have expected to find no
gender congruency effect, as found in the
Principle C conditions. Thus, it seems hard to
avoid the conclusion that the coreference interpretation is at least temporarily considered and subsequently rescinded.
Although the current results provide good evidence that a coreference interpretation is briefly
considered and then rescinded in the poka conditions, the results do not provide clear evidence
on the question of how actively the cataphoric
dependency is constructed in these conditions.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
389
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
A first possibility is that cataphoric dependency formation proceeded actively in the pokaconstraint conditions, just as in the no-constraint
conditions, leading to disruption upon encountering a gender-incongruent name (i.e., a gender
mismatch effect). If this was the case, then it must
be that the gender mismatch effect was overridden
in the reading-time profile by the immediately following disruption in the gender-congruent
condition, when the poka-constraint is triggered,
and the illicit interpretation had to be rescinded.
Note that this account entails that the processing
disruption caused by retraction of an illicit
pronoun–name dependency in the gendermatching poka condition must be stronger than
the processing disruption caused by encountering
an incongruent pronoun in the gender-mismatching
poka condition. This assumption could potentially
capture why no gender mismatch effect is observed
in these conditions. The assumption is reasonable,
in the respect that the gender mismatch effect
involves retraction of a prediction, whereas we are
claiming that the gender match effect reflects
retraction of a fully fledged cataphoric dependency.
A second possibility is that the construction of
cataphoric dependencies proceeds in a less active
fashion when a pronoun is introduced in a
poka-clause, and hence that a dependency is only
attempted after a gender-matching antecedent has
been encountered. The currently available data do
not allow us to adequately distinguish between
these accounts, but this does not undermine the
more important conclusion that a coreference
interpretation is fleetingly considered in the poka
conditions.
The finding that Principle C acts to block the
generation of coreference dependencies, whereas
the poka-constraint acts as a filter on dependencies
that are initially considered, naturally raises the
question of why the two constraints should
impact parsing so differently. The current findings
are consistent with a number of possible accounts
of this difference. It could reflect the fact that
Principle C applies to syntactic configurations
(Chomsky, 1981) whereas the poka-constraint is
a constraint that invokes discourse representations
(Avrutin & Reuland, 2004; Kazanina & Phillips,
390
2001; Reuland & Avrutin, 2005). Alternatively,
the difference could reflect the fact that Principle
C is a cross-linguistically robust constraint,
whereas the poka-constraint is a more idiosyncratic
property of Russian. In earlier work we found a
developmental dissociation between the two constraints: Russian children have already mastered
Principle C at age 3, whereas the poka-constraint
does not reliably constrain their interpretations
until age 5–6 (Kazanina & Phillips, 2001).
However, we suggest that the most likely cause
of the different online effects of the two constraints is the time course of information availability relevant to the two constraints. In the case
of Principle C the parser may determine that an
entire structural domain is irrelevant to the
search for an antecedent, and this can be ascertained in advance of any bottom-up information
about potential antecedents. This kind of precomputation is not possible in the case of the
poka-constraint. Recall that the poka-constraint
applies to pronoun –name sequences in which
both nominals have the semantic role of agent.
Although agent is by far the most common thematic role for a nominative subject, the thematic
role of the subject cannot be confirmed until the
predicate of the clause is identified. Therefore,
when faced with a sequence poka pronounNOM . . .
nameNOM the parser may continue to entertain
the possibility that the pronoun and the name
may be coreferential, because the thematic role of
the name is not yet confirmed. Only when the predicate is reached is it possible to confirm that the
name bears an agent role, and at that point the
poka-constraint becomes relevant, and the coreference relation must be rescinded, leading to
reading-time slowdown. This interpretation is
consistent with the time course of the gender
match effect, which in Experiment 1B appeared
at Region 13, slightly later than the gender mismatch effect in the no-constraint conditions. If
this is the appropriate account of why the pokaconstraint acts as a filter rather than as a constraint
on generation of antecedents then it is interesting
to ask why Russian speakers do not immediately
assume that a nominative subject is an agent, a
statistically likely outcome, and a plausible
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
consequence of parsing with superficial heuristics
(e.g., Towsend & Bever, 2001). However, the likelihood that a nominative subject NP is an agent
must be balanced against the likelihood that a
pronoun . . . name sequence of subject NPs involves
a coreference relation. Consequently, more
detailed analyses of Russian corpora would be
needed in order to determine whether speakers’
behaviour in our study genuinely ignores distributional constraints.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
CONCLUSION
A number of previous studies have investigated the
question of how grammatical constraints impact
the search for antecedents for anaphoric
expressions in online language processing. Either
constraints may act as constraints on the generation of candidate interpretations, which prevent
grammatically inaccessible antecedents from ever
being considered, or they may act as later filters
on interpretations, which serve to remove illicit
coreference relations only after they have been
initially considered. It is common for discussions
of this issue to assume that there is a uniform
answer to this question. However, results from
our within-subjects comparison of closely
matched constraints on backwards anaphora in
Russian indicate that a uniform answer may not
be possible. Russian speakers appear to implement
Principle C as a constraint on generation of
interpretations and the poka-constraint as a filter
on temporarily considered interpretations. This
contrast can be understood in terms of the detailed
time course of information that is relevant to
application of the two constraints.
Original manuscript received 1 September 2008
Accepted revision received 29 March 2009
First published online 6 July 2009
REFERENCES
Antonyuk, S., & Bailyn, J. (2008). Backward pronominalization in Russian: A syntactic account. Handout from
talk presented at the 17th meeting of Formal
Approaches to Slavic Linguistics (FASL-17). May,
2008, New Haven, CT, Yale University.
Aoshima, S., Yoshida, M., & Phillips, C. (2009).
Incremental processing of coreference and binding
in Japanese. Syntax, 12, 93– 134.
Avrutin, S., & Reuland, E. (2004). Backward anaphora
and tense interpretation. In L. Verbitskaya (Ed.),
Teoreticheskie problemy jazykoznanija [Theoretical
issues in language research]. St. Petersburg, Russia:
St. Petersburg University Press.
Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statistics using R. Cambridge, UK:
Cambridge University Press.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008).
Mixed-effects modeling with crossed random effects
for subjects and items. Journal of Memory and
Language, 59, 390– 412.
Badecker, W., & Straub, K. (2002). The processing role
of structural constraints on the interpretation of pronouns and anaphora. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 28,
748– 769.
Barss, A. (2003). Anaphora: A reference guide. Malden,
MA: Blackwell.
Büring, D. (2005). Binding theory. Cambridge, UK:
Cambridge University Press.
Chomsky, N. (1981). Lectures on government and
binding. Dordrecht, The Netherlands: Foris.
Clifton, C., Kennison, S. M., & Albrecht, J. E. (1997).
Reading the words her, his, him: Implications for
parsing principles based on frequency and structure.
Journal of Memory and Language, 36, 276– 292.
Cole, P., Hermon, G., & Huang, C.-T. J. (2000). Longdistance reflexives. New York: Academic Press.
Cowart, W., & Cairns, H. S. (1987). Evidence for an
anaphoric mechanism within syntactic processing:
Some reference relations defy semantic and pragmatic
constraints. Memory & Cognition, 15, 318–331.
Crain, S., & Fodor, J. D. (1985). How can grammars
help parsers. In D. Dowty, D. Kartunnen, &
A. M. Zwicky (Eds.), Natural language parsing:
Psycholinguistic, computational, and theoretical perspectives (pp. 94 – 128). Cambridge, UK: Cambridge
University Press.
Frazier, L., & Flores d’Arcais, G. B. (1989). Fillerdriven parsing: A study of gap filling in Dutch.
Journal of Memory and Language, 28, 331– 344.
Garnham, A. (2001). Mental models and the interpretation of anaphora. Hove, UK: Psychology Press.
Garnsey, S. M., Tanenhaus, M. K., & Chapman, R. M.
(1989). Evoked potentials and the study of sentence
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
391
Downloaded by [University Of Maryland] at 05:32 21 October 2011
KAZANINA AND PHILLIPS
comprehension. Journal of Psycholinguistic Research,
18, 51 – 60.
Gordon, P., & Hendrick, R. (1997). Intuitive knowledge of linguistic co-reference. Cognition, 62, 325–
370.
Harris, C. L., & Bates, E. A. (2002). Clausal backgrounding and pronominal reference: A functionalist
approach to c-command. Language and Cognitive
Processes, 17, 237–269.
Huang, Y. (2004). Anaphora and the pragmatics-syntax
interface. In L. R. Horn & G. Ward (Eds.), The
handbook of pragmatics (pp. 288– 314). Oxford, UK:
Blackwell.
Just, M. A., Carpenter, P. A., & Woolley, J. D. (1982).
Paradigms and processes in reading comprehension.
Journal of Experimental Psychology: General, 3, 228–238.
Kaan, E., Harris, A., Gibson, E., & Holcomb, P.
(2000). The P600 as an index of syntactic integration
difficulty. Language and Cognitive Processes, 15,
159 – 201.
Kaiser, E., & Trueswell, J. C. (2008). Interpreting pronouns and demonstratives in Finnish: Evidence for a
form-specific approach to reference resolution.
Language and Cognitive Processes, 23, 709– 748.
Kazanina, N. (2005). The acquisition and processing of
backwards anaphora. Unpublished PhD dissertation,
University of Maryland, College Park.
Kazanina, N., Lau, E., Lieberman, M., Yoshida, M., &
Phillips, C. (2007). The effect of syntactic constraints on the processing of backwards anaphora.
Journal of Memory and Language, 56, 384– 409.
Kazanina, N., & Phillips, C. (2001). Coreference in
child Russian: Distinguishing syntactic and discourse
constraints. In A. Do, L. Dominguez, & A. Johansen
(Eds.), Proceedings of the 25th Annual Boston
University Conference on Language Development
(pp. 413– 424). Somerville MA: Cascadilla Press.
Kennison, S. M. (2003). Comprehending the pronouns
her, him, and his: Implications for theories of referential processing. Journal of Memory and Language, 49,
335– 352.
Koster, J., & Reuland, E. (1992). Long-distance anaphora. Cambridge, UK: Cambridge University Press.
Lee, M.-W., & Williams, J. N. (2006). The role of grammatical constraints in intra-sentential pronoun resolution. Manuscript submitted for publication.
Matthiessen, C., & Thompson, S. A. (1988). The structure of discourse and “subordination”. In J. Haiman
392
& S. A. Thompson (Eds.), Clause combining in
grammar and discourse (pp. 275– 329). Amsterdam:
John Benjamins.
Nicol, J., & Swinney, D. (1989). The role of
structure in coreference assignment during sentence
comprehension. Journal of Psycholinguistic Research,
18, 5–19.
R Development Core Team (2009). R: A language and
environment for statistical computing. Vienna: R
Foundation for Statistical Computing.
Reinhart, T. (1983). Anaphora and semantic interpretation. London: Croom Helm.
Reinhart, T., & Reuland, E. (1993). Reflexivity.
Linguistic Inquiry, 24, 261 –321.
Reuland, E. (2001). Primitives of binding. Linguistic
Inquiry, 32, 439– 492.
Reuland, E., & Avrutin, S. (2005). Binding and beyond:
Issues in backward anaphora. In A. Branco,
T. McEnery, & R. Mitkov (Eds.), Anaphora processing: Linguistic, cognitive and computational modeling
(pp. 139– 162). Amsterdam: John Benjamins.
Runner, J. T., Sussman, R. S., & Tanenhaus, M. K.
(2006). Processing reflexives and pronouns
in picture noun phrases. Cognitive Science, 30,
193– 241.
Stowe, L. A. (1986). Parsing WH-constructions:
Evidence for on-line gap location. Language and
Cognitive Processes, 3, 227– 245.
Sturt, P. (2003). The time-course of the application of
binding constraints in reference resolution. Journal
of Memory and Language, 48, 542– 562.
Townsend, D. J., & Bever, T. G. (2001). Sentence comprehension: The integration of habits and rules.
Cambridge, MA: MIT Press.
Traxler, M. J., & Pickering, M. J. (1996). Plausibility
and the processing of unbounded dependencies: An
eye-tracking study. Journal of Memory and
Language, 35, 454– 475.
van Gompel, R. P. G., & Liversedge, S. P. (2003). The
influence of morphological information on cataphoric pronoun assignment. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 29,
128– 139.
van Hoek, K. (1997). Anaphora and conceptual structure.
Chicago: University of Chicago Press.
Xiang, M., Dillon, B., & Phillips, C. (2009). Illusory
licensing effects across dependency types: ERP evidence. Brain and Language, 108, 40 –55.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
APPENDIX A
Full list of stimuli (including comprehension questions) from the self-paced reading task,
Experiment 1
English translations of the experimental sentences are provided in order to give readers information about the scenarios tested. Note,
however, that the translations do not always preserve the Russian word order and exact phrasing, and no attempt is made to convey
the detailed aspectual force of the Russian predicates.
The experiment contained 24 sets the Principle C and poka conditions and 12 sets of no-constraint conditions:
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Principle C & poka conditions
No-constraint conditions
Condition a: Principle C, gender match
Condition b: Principle C, gender mismatch
Condition c: poka-constraint, gender match
Condition d: poka-constraint, gender mismatch
Condition a: no-constraint, gender match
Condition b: no-constraint, gender mismatch
Principle C and poka conditions
1a/b.
1c/d.
2a/b.
2c/d.
3a/b.
3c/d.
Qpslpm:lu qfrfe Rphefstcpn po sb9cam s rul eradpxfoopstj, qpla Narat/Amjoa q9tams>/q9tamas:
qpmuyjt: cji9 c Ardfotjou, Stas pqasams> c9icat: qpepirfojf u plruha<7jw.
Q9tams>/q9tamas: mj Narat/Amjoa qpmuyjt: cji9 c Braijmj<?
Since before Christmas he got rid of the jewellery while Marat/Alina attempted(m/f ) to secure visas to Argentina, Stas was afraid
to raise suspicions with acquaintances.
Qpslpm:lu qfrfe Rphefstcpn, qpla po sb9cam s rul eradpxfoopstj, Narat/Amjoa q9tams>/q9tamas:
qpmuyjt: cji9 c Ardfotjou, Stas pqasams> c9icat: qpepirfojf u plruha<7jw.
Q9tams>/q9tamas: mj Narat/Amjoa qpmuyjt: cji9 c Braijmj<?
Since before Christmas while he got rid of the jewellery, Marat/Alina attempted(m/f ) to secure visas to Argentina, Stas was
afraid to raise suspicions with acquaintances.
Qpslpm:lu uhf c qpmefo: po slruqumfiop jiuyam coutrfoo<< strulturu luqpma, qpla Njwajm/Tanara
efmam/efmama laraoeazo9f iarjspclj qpstrpklj, arwjtfltpr rassyjt9cam jnft: cs< ofpbwpejnu<
jovprnaxj< p wranf l lpoxu eo>.
Oaef>ms> mj arwjtfltpr ialpoyjt: sbpr ofpbwpejnpk jovprnaxjj ep lpoxa eo>?
Since already at noon he scrupulously studied the inner structure of the dome while Michael/Tamara made
(m/f ) pencil sketches of the construction, the architect counted on having all of the necessary information about the temple by
the end of the day.
Qpslpm:lu uhf c qpmefo:, qpla po slruqumfiop jiuyam coutrfoo<< strulturu luqpma, Njwajm/Tanara
efmam/efmama laraoeazo9f iarjspclj qpstrpklj, arwjtfltpr rassyjt9cam jnft: cs< ofpbwpejnu<
jovprnaxj< p wranf l lpoxu eo>.
Oaef>ms> mj arwjtfltpr ialpoyjt: sbpr ofpbwpejnpk jovprnaxjj ep lpoxa eo>?
Since already at noon while he scrupulously studied the inner structure of the dome, Michael/Tamara made(m/f ) pencil sketches of
the construction, the architect counted on having all of the necessary information about the temple by the end of the day.
Qpslpm:lu qfrfe oayampn sqfltalm> poa raidpcarjcama qp raxjj s pwraopk, qpla Rajsa/Sfrdfk rassahj
cama/rassahjcam qp nfstan c9splpqpstacmfoo9w dpstfk, Jooa of cjefma qrjlm<yjczfdps> s dfoframpn
laiusa.
Spstp>ms> mj raidpcpr s pwraopk qpsmf sqfltalm>?
Since before the beginning of the play she spoke with security on the portable radio while Raisa/Sergej sat(m/f ) VIP guests in their
seats, Inna didn’t notice the incident that happened to the general.
Qpslpm:lu qfrfe oayampn sqfltalm>, qpla poa raidpcarjcama qp raxjj s pwraopk, Rajsa/Sfrdfk rassahj
cama/rassahjcam qp nfstan c9splpqpstacmfoo9w dpstfk, Jooa b9ma pyfo: bmadpearoa fk/fnu ia qpnp7:.
Spstp>ms> mj raidpcpr s pwraopk qpsmf sqfltalm>?
Since before the beginning of the play while she spoke with security on the portable radio, Raisa/Sergej sat(m/f ) VIP guests in their
seats, Inna was very grateful to her/him for help.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
393
KAZANINA AND PHILLIPS
4a/b.
4c/d.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
5a/b.
5c/d.
6a/b.
6c/d.
7a/b.
7c/d.
8a/b.
8c/d.
394
Qpslpm:lu qfrfe ;vjrpn poa qrpsnatrjcama tflst9 sppb7fojk, qpla Narjoa/Eaojjm drjnjrpcamas:/
drjnjrpcams> l oayamu s:fnpl, Ip> qfrcpk uioama sfosaxjpoou< opcpst:. Uioama mj Ip> sfosaxjpoou<
opcpst: ep oayama ;vjra?
Since before the broadcast she looked through the news texts while Marina/Danil put(m/f ) on make up for the shoot, Zoja was the
first one to learn about the sensational news.
Qpslpm:lu qfrfe ;vjrpn, qpla poa qrpsnatrjcama tflst9 sppb7fojk, Narjoa/Eaojjm drjnjrpcamas:/
drjnjrpcams> l oayamu s:fnpl, Ip> sana pqrfefmjma qpr>epl rfqprtahfk c c9quslf.
B9m mj pqrfefmfo qpr>epl opcpstfk Ipfk?
Since before the broadcast while she looked through the news texts, Marina/Danil put(m/f) on make up for the shoot, Zoja figured out
the order of the reports in the program by herself.
Qpslpm:lu cyfra eofn po qpelm<yam jinfrjtfmj eacmfoj>, qpla Artur/Lmara oastrajcam/oastrajcama
sqfxjam:ou< caluunou< lanfru, Cjtamjk b9m ucfrfo c qpmopk sjowrpojiaxjj ustaopclj.
Spnofcams> mj Cjtamjk c sjowrpojiaxjj ustaopclj?
Since yesterday during the day he connected the pressure meter while Arthur/Clara tuned(m/f ) the special vacuum camera, Valerij was
certain that the equipment was fully synchronized.
Qpslpm:lu cyfra eofn, qpla po qpelm<yam jinfrjtfmj eacmfoj>, Artur/Lmara oastrajcam/oastrajcama
sqfxjam:ou< caluunou< lanfru, Camfrjk oaef>ms> oamaejt: pbpruepcaojf ep oastuqmfoj> sunfrfl.
Qptfr>m mj Camfrjk oaefheu oastrpjt: pbpruepcaojf ep tfnopt9?
Since yesterday during the day while he connected the pressure meter Arthur/Clara tuned(m/f) the special vacuum camera, Valerij
hoped to fix the equipment before dusk.
Qpslpm:lu cs< ofefm< qfrfe qaro9n vjoampn po bfsqfyop iajdr9cam s nopdpyjsmfoo9nj qplmpoojxanj,
qpla Csfcpmpe/Lrjstjoa ptrabat9cam/ptrabat9cama oa lprtf lm<yfc9f uear9, Enjtrjk pt9dram c vjoa
m:opn natyf ioayjtfm:op wuhf scpfdp qartofra/scpfk qartofrzj.
Jdram mj Enjtrjk c joejcjeuam:opn vjoamf?
Since all week before the doubles final he carelessly flirted with numerous fans, while Vsevolod/Kristina worked(m/f ) out key strokes
on the court, Dmitri played far worse in the final than his partner.
Qpslpm:lu cs< ofefm< qfrfe qaro9n vjoampn, qpla po bfsqfyop iajdr9cam s nopdpyjsmfoo9nj qplmpooj
xanj, Csfcpmpe/Lrjstjoa ptrabat9cam/ptrabat9cama oa lprtf lm<yfc9f uear9, Enjtrjk of npd of
qrjioat: clmae scpfdp qartofra/scpfk qartofrzj rfza<7jn c jw qpbfef.
Jdram mj Enjtrjk c joejcjeuam:opn vjoamf?
Since all week before the doubles final while he carelessly flirted with numerous fans Vsevolod/Kristina worked(m/f ) out key strokes on the
court, Dmitri had to acknowledge his partner’s contribution to their victory.
Qpslpm:lu qpsmf lpotrpm:opk poa qrpsnatrjcama uyfojyfsljf tftraej, qpla Pm:da/Bprjs ccpejma/
ccpejm c lpnq:<tfr statjstjlu qp pzjblan, uyjtfm:ojxa npdma sracojcat: rfium:tat9 nfheu lmassanj.
Ccpejma/ccpejm mj Pm:da/Bprjs statjstjlu qp pzjblan c lpnq:<tfr?
Since before the exam she looked through students’ notebooks while Olga/Boris entered(f/m) data about mistakes into the computer, the
teacher was able to compare results between classes.
Qpslpm:lu qpsmf lpotrpm:opk, qpla poa qrpsnatrjcama uyfojyfsljf tftraej, Pm:da/Bprjs ccpejma/
ccpejm c lpnq:<tfr statjstjlu qp pzjblan, uyjtfm:ojxa ioama qperpbo9f rfium:tat9 sraiu qp
plpoyaojj qrpcfrlj.
Ccpejma/ccpejm mj Pm:da/Bprjs statjstjlu qp pzjblan c lpnq:<tfr?
Since before the exam while she looked through students’ notebooks Olga/Boris entered(f/m) data about mistakes into the computer,
the teacher knew the detailed results immediately after she finished checking the papers.
Wpt> c qrpzmpn dpeu po cqfrc9f tfstjrpcam opc9k cje ejft9, qpla 6rpsmac/Carcara dptpcjms>/
dptpcjmas: l ptcftstcfoopnu ;taqu yfnqjpoata njra, Dmfb bfi pqasfojk coferjm scpj opczfstca c
swfnu qjtaoj> sqprtsnfoa/sqprtsnfolj.
Jsq9t9cam mj Dmfb pqasfoj> ptopsjtfm:op opcpk ejft9?
Even though in the past year he tested a new type of diet for the first time while Yaroslav/Varvara prepared(m/f ) for the crucial
round of the world championship, Gleb fearlessly introduced his own innovations into the competitor’s diet.
Wpt> c qrpzmpn dpeu, qpla po cqfrc9f tfstjrpcam opc9k cje ejft9, 6rpsmac/Carcara dptpcjms>/dptpcjmas:
l ptcftstcfoopnu ;taqu yfnqjpoata njra, Dmfb bfi pqasfojk coferjm scpj opczfstca c swfnu qjtaoj>
sqprtsnfoa/sqprtsnfolj.
Qpbp>ms> mj Dmfb jsq9tat: opc9k cje ejft9 qfrfe yfnqjpoatpn njra?
Even though in the past year while he tested a new type of diet for the first time Yaroslav/Varvara prepared(m/f ) for the crucial round of
the world championship, Gleb fearlessly introduced his own innovations into the competitor’s diet.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
9a/b.
9c/d.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
10a/b.
10c/d.
11a/b.
11c/d.
12a/b.
12c/d.
13a/b.
13c/d.
14a/b.
Wpt> c >ocarf poa dptpcjmas: l lcamjvjlaxjpoo9n ;lianfoan, qpla Amma/Jcao pte9wama/pte9wam oa
Laoarsljw pstrpcaw, Cjla of sftpcama oa sue:bu.
Pte9wama/pte9wam mj Amma/Jcao oa Laoaraw c acdustf?
Even though in January she prepared for the qualifying exams while Alla/Ivan relaxed(f/m) in the Canary Islands, Vika did not
complain about life.
Wpt> c >ocarf, qpla poa dptpcjmas: l lcamjvjlaxjpoo9n ;lianfoan, Amma/Jcao pte9wama/pte9wam oa
Laoarsljw pstrpcaw, Cjla of sftpcama oa sue:bu.
Pte9wama/pte9wam mj Amma/Jcao oa Laoaraw c acdustf?
Even though in January while she prepared for the qualifying exams Alla/Ivan relaxed(f/m) in the Canary Islands, Vika did not
complain about life.
Wpt> c cpslrfsfo:f poa uyjma bjmft9 qp vjijlf, qpla Car>/Cpca smuzama/smuzam qr>npk raejprfqpr
tah s ‘Fcrpcjefoj>’, Rjta unuer>mas: of pbra7at: cojnaoj> oa qrpjswpe>7ff.
Traosmjrpcamj mj ‘Fcrpcjefojf’ qp raejp c qr>npn ;vjrf?
Even though on Sunday she studied the physics exam questions while Varja/Vova listened (f/m) to radio coverage of Eurovision, Rita
managed not to pay attention to what was going on.
Wpt> c cpslrfsfo:f, qpla poa uyjma bjmft9 qp vjijlf, Car>/Cpca smuzama/smuzam qr>npk raejprfqpr
tah s
‘Fcrpcjefoj>’, Rjta unuer>mas: of pbra7at: cojnaoj> oa qrpjswpe>7ff.
Traosmjrpcamj mj ‘Fcrpcjefojf’ qp raejp c qr>npn ;vjrf?
Even though on Sunday while she studied the physics exam questions Varja/Vova listened (f/m) to radio coverage of Eurovision, Rita
managed not to pay attention to what was going on.
Wpt> qpytj csf utrp po jslam qrfestacjtfmfk rfelpdp cjea dusfojx9, qpla Fdpr/Naza sojnam/sojnama oa
lanfru oajbpmff lrasjc9f udpmlj qarla, Arlaejk tal j of oazfm ouho9k fnu ;lifnqm>r.
Qracea mj, ytp qpjslj dusfojx9 cfmjs: cfyfrpn?
Even though almost all morning he searched for examples of a rare caterpillar species while Egor/Masha photographed(m/f) the
more beautiful corners of the park, Arkadi still did not find the specimen he was looking for.
Wpt> qpytj csf utrp, qpla po jslam ;lifnqm>r rfelpdp cjea dusfojx9, Fdpr/Naza sojnam/sojnama oa
lanfru oajbpmff lrasjc9f udpmlj qarla, Arlaejk qfrfhjcam, ytp nam:yjl/efcuzla sluyaft.
Qracea mj, ytp qpjslj dusfojx9 cfmjs: cfyfrpn?
Even though almost all morning while he searched for examples of a rare caterpillar species Egor/Masha photographed(m/f) the
more beautiful corners of the park, Arkadij was worried that the boy/girl was bored.
Wpt> lahepf cpslrfsfo:f poa qjma utrfoojk lpvf, qpla Pm>/5ra rfzama/rfzam lalpkojbue: ucmflatf
m:o9k lrpsscpre, Narjoa ojlpdea of qrfemadama qpnpy:.
Qracea mj, ytp Narjoa ojlpdea of qrfemadama scpfk qpnp7j c raidaelf lrpsscprepc?
Even though every Sunday she drank morning coffee while Olja/Jura solved(f/m) some fascinating crossword puzzle, Marina
never offered help.
Wpt> lahepf cpslrfsfo:f, qpla poa qjma utrfoojk lpvf, Pm>/5ra rfzama/rfzam rfzama lalpk-ojbue:
ucmflatfm:o9k lrpsscpre, Narjoa ojlpdea of qrjojnama uyastj> c raidaelf cpqrpspc.
Qracea mj, ytp Narjoa ojlpdea of qrjojnama uyastj> c raidaelf lrpsscprepc?
Even though every Sunday while she drank morning coffee Olja/Jura solved(f/m) some fascinating crossword puzzle, Marina
never helped out with the crossword.
Qpslpm:lu c nadaijof po bpmtam s sjnqatjyo9nj lpoejtfrzanj, qpla Nalsjn/Oataza qrpbpcam/
qrpbpcama raio9f cje9 qjrpho9w, Ejna qpmuyjm pdrpnopf uepcpm:stcjf pt qpwpea ia qpluqlanj.
Qracea mj, ytp c nadaijof rabptamj sjnqatjyo9f lpoejtfrzj?
Since in the store he bantered with the pretty bakers while Maksim/Natasha tasted(m/f) different types of pastries, Dima greatly
enjoyed their shopping trips.
Qpslpm:lu c nadaijof, qpla po bpmtam s sjnqatjyo9nj lpoejtfrzanj, Nalsjn/Oataza qrpbpcam/
qrpbpcama raio9f cje9 qjrpho9w, Ejna iawptfm tphf qpqrpbpcat: ytp-ojbue: ji smaelpdp.
Qracea mj, ytp c nadaijof nphop b9mp qpqrpbpcat: c9qfylu?
Since in the store while he bantered with pretty bakers Maksim/Natasha tasted(m/f) different types of pastries, Dima also expressed
a desire to try some pastry.
Wpt> uhf c qpmzfstpdp poa ulrazama ifmfo:< samat9, qpla Damjoa/Sfrfha sqfzop oalr9cama/oalr9cam oa
stpm, wpi>kla tal j of usqfma ialpoyjt: qrjdptpcmfoj> ep qrjwpea dpstfk.
Ulrazamjs: mj samat9 mjnpopn?
Even though already at half past five she decorated the salads with greens while Galina/Serezha hastily set(f/m) the table, the hostess did
not manage to finish the preparations before the guests arrived.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
395
KAZANINA AND PHILLIPS
14c/d.
15a/b.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
15c/d.
16a/b.
16c/d.
17a/b.
17c/d.
18a/b.
18c/d.
19a/b.
19c/d.
396
Wpt> uhf c qpmzfstpdp, qpla poa ulrazama ifmfo:< samat9, Damjoa/Sfrfha sqfzop oalr9cama/oalr9cam oa
stpm, wpi>kla of oaef>mas: ialpoyjt: qrjdptpcmfoj> ep qrjwpea dpstfk.
Ulrazamjs: mj samat9 mjnpopn?
Even though already at half past five while she decorated the salads with greens Galina/Serezha hastily set(f/m) the table, the hostess did
not hope to finish the preparations before the guests arrived.
Wpt> cs< subbptu poa qpmpma raibuzfcaczjfs> spro>lj, qpla Scfta/Caejn srfiama/srfiam suwjf cftlj s
lustpc, Camfotjoa, c ptmjyjf pt qperudj/nuha, spcfrzfoop of yucstcpcama ustampstj.
Utpnjma mj Camfotjou rabpta c saeu?
Even though all Saturday she dug up the out-of-control weeds while Sveta/Vadim cut(f/m) the dry branches from the bushes,
Valentina, unlike her friend(f)/husband, felt no tiredness.
Wpt> cs< subbptu, qpla poa qpmpma raibuzfcaczjfs> spro>lj, Scfta/Caejn srfiama/srfiam suwjf cftlj s
lustpc, Camfotjoa syjtama ubprlu saea qpmopst:< scpfk iasmudpk.
B9ma mj Camfotjoa pb8fltjcoa c scpfk pxfolf?
Even though all Saturday while she dug up the out-of-control weeds Sveta/Vadim cut(f/m) the dry branches from the bushes,
Valentina considered the garden clean-up to be solely her achievement.
Qpslpm:lu cfs: cfyfr poa spstacm>ma lcartam:o9k ptyft, qpla Mjea/Cas> snptrfma/snptrfm qpctpr
‘Rphefstcfosljw cstrfy’, Narjoa syjtama sfb> cqracf of dptpcjt: uhjo.
Snptrfma/snptrfm mj Mjea/Cas> qpctpr ‘Rphefstcfosljw cstrfy?
Since all evening she prepared the quarterly report while Lida/Vasja watched(f/m) the rerun of the “Christmas meetings”, Marina
considered herself justified not to prepare dinner.
Qpslpm:lu cfs: cfyfr, qpla poa spstacm>ma lcartam:o9k ptyft, Mjea/Cas> snptrfma/snptrfm qpctpr
‘Rphefstcfosljw cstrfy’, Narjoa of npdma qpmopst:< spsrfeptpyjt:s> oa rabptf.
Snptrfma/snptrfm mj Mjea/Cas> qpctpr ‘Rphefstcfosljw cstrfy?
Since all evening while she prepared the quarterly report, Lida/Vasja watched(f/m) the rerun of the “Christmas meetings”, Marina
failed to concentrate fully on her work.
Wpt> c rfstpraof poa lurjma oa tfrassf, qpla Amjoa/Bprjs pbsuheama/pbsuheam s pvjxjaotpn qrfmfstj
>qposlpk luwoj, Dam> sm9zama csf eftamj jw raidpcpra.
Pbsuheama/pbsuheam mj Amjoa/Bprjs s pvjxjaotpn jtam:>oslu< luwo<?
Even though in the restaurant she smoked on the terrace while Alina/Boris discussed(f/m) the charms of Japanese cuisine with the
waiter, Galja caught their conversation in detail.
Wpt> c rfstpraof, qpla poa lurjma oa tfrassf, Amjoa/Bprjs pbsuheama/pbsuheam s pvjxjaotpn qrfmfstj
>qposlpk luwoj, Dam> oastp>ma oa ialaif jslm<yjtfm:op fcrpqfksljw bm<e.
Pbsuheama/pbsuheam mj Amjoa/Bprjs s pvjxjaotpn jtam:>oslu< luwo<?
Even though in the restaurant while she smoked on the terrace Alina/Boris discussed(f/m) the charms of Japanese cuisine with the
waiter, Galja insisted on an order consisting exclusively of European dishes.
Qpslpm:lu c sanpmftf poa raszjcama bmfstlanj lpst<n, qpla Amfoa/Drjza iauyjcama/iauyjcam oajiust:
smpca vjoam:opk qfsoj, Haooa staramas: of ptcmflat: epy:/s9oa scpjnj qrps:banj j rassqrpsanj.
Qracea mj, ytp Haooa staramas: of bfsqplpjt: epy:/s9oa?
Since on the plane she embroidered the suit with sequins while Alena/Gregory learned the words of the final song by heart, Zhanna
tried not to distract her daughter/son with her requests and questions.
Qpslpm:lu c sanpmftf, qpla poa raszjcama bmfstlanj lpst<n, Amfoa/Drjza iauyjcama/iauyjcam
oajiust: smpca vjoam:opk qfsoj, Haooa ofcpm:op c9uyjma tflst cnfstf s epyfr:</s9opn.
C9uyjma mj Haooa tflst tflst vjoam:opk qfsoj?
Since on the plane while she embroidered the suit with sequins Alena/Gregory learned the words of the final song by heart, Zhanna
reluctantly leant the lyrics together with her daughter/son.
Qpslpm:lu csf utrp po q9mfspsjm lpcr9 c mpehjj, qpla Efojs/Ls<za pbicaojcam/pbicaojcama eruifk
ptopsjtfm:op qrfestp>7fk cstrfyj, Sfnfo oj laqmj of sphamfm p rfzfojj cstrftjt:s> u ojw epna.
B9mj mj qpecfrdout9 yjstlf lpcr9 c mpehjj?
Since all morning he vacuumed carpets on the balcony while Denis/Ksjusha telephoned(m/f) friends about the upcoming meeting,
Simon did not regret at all the decision to meet at their place.
Qpslpm:lu csf utrp, qpla po q9mfspsjm lpcr9 c mpehjj, Efojs/Ls<za pbicaojcam/pbicaojcama eruifk
ptopsjtfm:op qrfestp>7fk cstrfyj, Sfnfo qmptop ialr9m csf ecfrj j starams> of zunft:.
B9mj mj qpecfrdout9 yjstlf lpcr9 c mpehjj?
Since all morning while he vacuumed carpets on the balcony Denis/Ksjusha telephoned(m/f) friends about the upcoming meeting,
Simon tightly closed all the doors and tried not to make any noise.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
20a/b.
20c/d.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
21a/b.
21c/d.
22a/b.
22c/d.
23a/b.
23c/d.
24a/b.
24c/d.
Wpt> cfs: efo: po c9qjs9cam psopco9f xjtat9, qpla Jm:>/Mjoa qjsam/qjsama oaioayfoo9k oa yftcfrd
eplmae, Ljrjmm of wptfm cojlat: c sut: fdp/fg rabpt9.
Cojlam mj Ljrjmm c sut: eplmaea?
Even though all day he wrote out the main citations while Ilya/Lina wrote(m/f) the report that was due on Thursday, Kirill did
not want to grasp the essence of his/her work.
Wpt> cfs: efo:, qpla po c9qjs9cam psopco9f xjtat9, Jm:>/Mjoa qjsam/qjsama oaioayfoo9k oa yftcfrd
eplmae, Ljrjmm of wptfm cojlat: c sut: fdp/fg rabpt9.
Cojlam mj Ljrjmm c sut: eplmaea?
Even though all day while he wrote out the main citations Ilya/Lina wrote(m/f) the report that was due on Thursday, Kirill did
not want to grasp the essence of his/her work.
Wpt> eca eo> oaiae po pbrabat9cam spbraoo9f raoff eaoo9f, qpla Jdpr:/Scfta qrpcpejm/qrpcpejma
qpctpro9k lpotrpm:o9k ;lsqfrjnfot, Dfooaejk of iawptfm efmjt:s> s ojn/ofk rfium:tatanj.
Qracea mj, ytp lpotrpm:o9k ;lsqfrjnfot qrpcpejms> qpctprop?
Even though two days ago he processed the data collected previously while Igor/Sveta performed the control experiment, Gennady
did not want to share the results with him/her.
Wpt> eca eo> oaiae, qpla po pbrabat9cam spbraoo9f raoff eaoo9f, Jdpr:/Scfta qrpcpejm/qrpcpejma
qpctpro9k lpotrpm:o9k ;lsqfrjnfot, Dfooaejk of iawptfm efmjt:s> s ojn/ofk rfium:tatanj.
Qracea mj, ytp lpotrpm:o9k ;lsqfrjnfot qrpcpejms> qpctprop?
Even though two days ago while he processed the data collected previously Igor/Sveta performed the control experiment, Gennady
did not want to share the results with him/her.
Qpslpm:lu c qrpzmu< q>tojxu po qrpcftrjcam rabpyff qpnf7fojf, qpla 6lpc/4mma rasqfyat9cam/rasq
fyat9cama lpqjj qraieojyo9w rflmano9w mjstpcpl, Lpst> cjojm sfb> c qrpstuef lpmmfdj.
Qracea mj, ytp 6lpc/4mma rasqfyat9cam/rasqfyat9cama lpqjj buwdamtfrsljw vprn?
Since last Friday he aired out the work space while Yakov/Ella printed(m/f) copies of the holiday advertisement flyers, Kostja
blamed himself for his colleague’s illness.
Qpslpm:lu c qrpzmu< q>tojxu, qpla po qrpcftrjcam rabpyff qpnf7fojf, 6lpc/4mma rasqfyat9cam/rasq
fyat9cama lpqjj qraieojyo9w rflmano9w mjstpcpl, Lpst> oalpofx qpioalpnjms> s opc9n sptrueojlpn/
opcpk sptrueojxfk.
Qracea mj, ytp 6lpc/4mma rasqfyat9cam/rasqfyat9cama lpqjj buwdamtfrsljw vprn?
Since last Friday while he aired out the work space Yakov/Ella printed(m/f) copies of the holiday advertisement flyers, Kostja
finally met his new coworker (m/f).
Wpt> s utra po efmam ubprlu c lcartjrf, qpla Lpm>/Ijoa smuzam/smuzama opc9k am:bpn ‘Amjs9’, Jcao of
stam uqrflat: fdp/fg oj c yfn.
Qracea mj, ytp Jcao oj c yfn of uqrfloum Lpm</Ijou?
Even though since the morning he cleaned the apartment while Kolja/Zina listened (m/f) to the new album by «Alisa», Ivan did
not reproach him/her for anything.
Wpt> s utra, qpla po efmam ubprlu c lcartjrf, Lpm>/Ijoa smuzam/smuzama opc9k am:bpn ‘Amjs9’, Jcao of
stam uqrflat: fdp/fg oj c yfn.
Qracea mj, ytp Jcao oj c yfn of uqrfloum Lpm</Ijou?
Even though since the morning while he cleaned the apartment Kolja/Zina listened (m/f) to the new album by «Alisa», Ivan did
not reproach him/her for anything.
Wpt> c qarlf poa uo9mp dr9ima t9lcfoo9f sfnfylj, qpla Cfra/Cjt> yjtama/yjtam pb8>cmfoj> oa stfoef,
5m> of wptfma tprpqjt: qperudu/eruda.
Qracea mj, ytp Cfra/Cjt> yjtama/yjtam pb8>cmfoj> c huroamf?
Even though in the park she glumly munched on pumpkin seeds while Vera/Vitja read(f/m) ads on the stand, Julia did not want to
hurry her/him up.
Wpt> c qarlf, qpla poa uo9mp dr9ima t9lcfoo9f sfnfylj, Cfra/Cjt> yjtama/yjtam pb8>cmfoj> oa stfoef,
5m> of iawptfma qrjspfejojt:s> l ofk/ofnu.
Qracea mj, ytp Cfra/Cjt> yjtama/yjtam pb8>cmfoj> c huroamf?
Even though in the park while she glumly munched on pumpkin seeds Vera/Vitja read(f/m) ads on the stand, Julia did not want to
join her/him.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
397
KAZANINA AND PHILLIPS
No-constraint conditions
1a/b.
2a/b.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
3a/b.
4a/b.
5a/b.
6a/b.
7a/b.
8a/b.
9a/b.
398
Wpt> ep tpdp lal poa ptlr9ma spbstcfoou< vjrnu, Aooa/Ejna wprpzp iarabat9cama/iarabat9cam
qfrfcpeanj c Mulpkmf, Ejna/Aooa of spnofcams> c qracjm:opstj fg rfzfoj>/of spnofcamas: c ofpbw
pejnpstj qfrfwpea nuha c sfnfko9k bjiofs.
B9ma/b9m mj Aooa/Ejna ofepcpm:oa/ofepcpmfo iarqmatpk c Mulpkmf?
Even though before she started her own company Anna/Dima earned(f/m) good money translating for Lukoil, Dima/Anna did
not doubt {the correctness of her decision/the necessity of her husband’s move into the family business}.
Wpt> qpsmf tpdp lal po ialpoyjm dfofram:ou< rfqftjxj<, Cmaejnjr/Lrjstjoa b9m/b9ma qpmopst:<
epcpmfo sqfltalmfn, Lrjstjoa/ Cmaejnjr of raiefm>ma/raiefm>m fdp/fg ;otuijaina.
Raimjyamjs: mj nofoj> Lrjstjo9 j Cmaejnjra c ptopzfojj sqfltalm>?
Even though after he finished the dress rehearsal Vladimir/Kristina was(m/f) completely satisfied with the play, Kristina/
Vladimir did not share(f/m) her/his enthusiasm.
Wpt> qpsmf tpdp lal poa oaqjsama ialaiaoou< stat:<, Camfotjoa/Amflsaoer ofslpm:lp rai qracjma/
qracjm tflst, Amflsaoer csf-talj b9m ofepcpmfo ff qpeayfk natfrjama/Camfotjoa bpm:zf csfdp dprej
mas: scpjn qfrcpoayam:o9n carjaotpn.
Qracjma/qracjm mj Camfotjoa/Amflsaoer qfrcpoayam:o9k tflst stat:j?
Even though after she wrote the commissioned article Valentina/Aleksandr corrected the text a few times, {Aleksandr nonetheless
was dissatisfied with her presentation of the material/Valentina was most proud of her initial version}.
Wpt> ep tpdp lal po pqubmjlpcam pbipr c Mjtfraturopk Daiftf, Dacrjjm/Oaefhea b9m/b9ma ofjicfstfo/
ofjicfstoa c qjsatfm:sljw lrudaw, Oaefhea of spnofcamas: c qp>cmfojj jotfrfsa l ofnu qpsmf c9wpea
pbipra/Dacrjjm of spnofcams> c qp>cmfojj jotfrfsa l ofk qpsmf c9wpea fdp pbipra.
B9m/b9ma mj Dacrjjm/Oaefhea zjrplp jicfstfo/jicfstoa qjsatfm>n ep qubmjlaxjj c Mjtfraturopk
Daiftf?
Even though before he published the review in Literaturnaja Gazeta Gavriil/Nadezhda was(m/f) unknown(m/f) in literary circles,
{Nadezhda did not doubt that there would be much interest in him after the review appeared/Gavriil did not doubt that there would
be much interest in her after the review appeared}.
Wpt> qpsmf tpdp lal po c9jdram ijnojf sprfcopcaoj>, Qacfm/Oast> ofecusn9smfoop oanfloum/oan
flouma oa uyastjf c Pmjnqjaef, Oast> ptlaiamas: pt m<b9w lpnnfotarjfc ptopsjtfm:op scpfdp qpepq
fyopdp/Qacfm ptlaiams> pt m<b9w lpnnfotarjfc ptopsjtfm:op fg c9slai9caoj>.
Oanfloum/oanflouma mj Qacfm/Oast> oa uyastjf c Pmjnqjaef?
Even though after he won the winter competitions Pavel/Nastja unambiguously hinted(m/f) at participating in the Olympics,
{Nastja refused to make any comments about her protégé/Pavel refused to comment upon her statement}.
Wpt> ep tpdp lal poa cpzma c spstac sbpropk, Oataza/Njwajm b9ma/b9m spcfrzfoop ofjicfstoa/
ofjicfstfo sqfxjamjstan, Njwajm wprpzp qpojnam ioayjnpst: ff usqfwa/Oataza ojlpdea of spnofcamas:
c fdp c9ea<7jws> trfofrsljw sqpspbopst>w.
B9ma/b9m mj Oataza/Njwajm ioanfojtpk/ioanfojt9n?
Even though before she entered the team line-up Natasha/Mixail was(f/m) completely unknown(f/m) to aficionados, {Michael well
understood the significance of her success/Natasha never doubted his outstanding coaching talent}.
Qpslpm:lu qpsmf tpdp lal po qfrffwam rabptat: oa sfcfr, 5rjk/Lat> stam/stama qpmuyat: aopojno9f
udrpi9, Lat> csfnj sjmanj q9tamas: udpcprjt: fdp cfrout:s> c efrfco</5rjk rfzjm iabrat: sfstru l
sfbf c Oprjm:sl.
Q9tamas: mj Lat> udpcprjt: 5rj> ufwat: s sfcfra?/Qmaojrpcam mj 5rjk iabrat: l sfbf sfstru?
Since after he moved to the north for work Yurij/Katja started(m/f) receiving anonymous threats, {Katja tried everything possible to
convince him to return to the village/Yurij decided to move his sister to his place in Norilsk}.
Qpslpm:lu ep tpdp lal po of qpmuyjm qfyat:, Ejna/Ojoa of qpeqjs9cam/qpeqjs9cama ojlaljw bunad,
Ojoa b9ma c9ouhefoa ptmphjt: ialm<yfojf lpotraltpc/Ejna b9m c9ouhefo ptmphjt: qfrfdpcpr9 p
ialm<yfojj lpotralta.
Qracea mj, ytp csf lpotralt9 b9mj qpeqjsao9 bfi ptmadatfm:stc?/Qracea mj, ytp qfrfdpcpr9 b9mj
qrpcfefo9 c jioayam:op qmaojrufnpf crfn>?
Since until he received the stamp Dima/Nina did not sign(m/f) any papers, {Nina was forced to postpone the signing of the contracts/Dima was forced to postpone the contract-related negotiations}.
Qpslpm:lu ep tpdp lal poa of raipbramas: cp csfw eplunfotaw, Plsaoa/Ojljta of qrpcpejma/qrpcpejm
qfrfdpcprpc s lmjfotanj, Ojljta of npd sraiu pxfojt: ff efmpc9f layfstca/Plsaoa of npdma rassyjt9
cat: oa qrfnj<.
Trfbpcamps: mj Plsaof crfn> oa pioalpnmfojf s eplunfotanj?/Cfrop mj, ytp Plsaoa of oaef>mas: oa
qrfnj<?
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
PROCESSING RUSSIAN CATAPHORA
10a/b.
11a/b.
Downloaded by [University Of Maryland] at 05:32 21 October 2011
12a/b.
Since until she understood all the documents Oksana/Nikita did not conduct(f/m) any negotiations with clients, {Nikita could not
immediately appreciate her professional qualities/Oksana could not expect a bonus}.
Qpslpm:lu ep tpdp lal po ustrpjms> c pwraoopf adfostcp, Ojlpmak/Marjsa qpstp>oop qfrfslai9cam/
qfrfslai9cama qrpyjtaoo9f eftfltjc9, Marjsa syjtama, ytp opca> rabpta qrjefts> fnu qp euzf/
Ojlpmak oaef>ms> sraijt: qperudu iawcat9ca<7jnj smuya>nj ji scpfk qraltjlj.
Qracea mj, ytp Marjsa ustrpjmas: oa rabptu c pwraoopf adfostcp?
Since before he settled into a job with a security agency Nikolai/Larisa constantly repeated(m/f) the detective stories he/she had
read, {Larisa thought that the new job would suit him/Nikolai hoped to impress his girlfriend with fascinating cases from his own
work}.
Qpslpm:lu qpsmf tpdp lal poa qfrfzma oa erudu< rabptu, Eaza/Pmfd stama/stam iabjrat: eftfk ji
saejla, Pmfd/Eaza b9m/b9ma pyfo: rae/raea qp>cjczfnus> u ofdp/ofg scpbpeopnu crfnfoj.
Iabjrama mj Eaza eftfk ji saea?/B9ma mj Eaza pdpryfoa qp>cjczfns> scpbpeo9n crfnfofn?
Since before she transferred to a new job Dasha/Oleg started(f/m) to pick-up children from kindergarten, Oleg/Dasha was (m/f)
very happy(m/f) with the free time available to him/her.
Qpslpm:lu ep tpdp lal poa oayama iaojnat:s> matjopanfrjlaosljnj taoxanj, Jrjoa/Aotpo
qrpvfssjpoam:op iaojnamas:/iaojnams> lmassjyfsljn bamftpn, Aotpo sraiu hf qp epstpjostcu pxfojm
bueu7u< qartofrzu/Jrjoa s bpm:zjn truepn udpcprjma fdp stat: ff qartofrpn.
Qracea mj, ytp Jrjoa/Aotpo iaojnamas:/iaojnams> bamftpn?
Since before she started to be involved in Latin-American dances Irina/Anton danced(f/m) classical ballet professionally, {Anton
immediately recognized his future partner’s talents/Irina convinced him with great difficulty to be her partner}.
APPENDIX B
Mean reading times and standard error (in ms) for each region in each condition in the online
Experiment 1b
Fixed effect(s) for the best fitting model for each region are shown in the last column with “—” representing the model with no fixed
factors other than the intercept. All models had participants and items as random factors.
No-constraint conditions
No constraint
Region
Gender match
Gender mismatch
Best fitting model
1
2
3
4
5
6
7
8
9 (2nd subject)
10
11
12
13
14 (3rd subject)
15
16
17
481.8 (13.8)
419.1 (13.6)
416.6 (10.5)
404.6 (11.1)
398.8 (10.5)
472.9 (16.8)
523.3 (22.3)
551.1 (23.5)
533 (19.3)
487 (16.9)
541.6 (20.1)
546.6 (21.9)
542.1 (20.1)
568.2 (20.4)
469.2 (13.8)
461.1 (15.7)
476 (17.3)
480.8 (15.8)
424.6 (14.1)
414.1 (11.3)
393.2 (10.2)
412.6 (13.9)
473.8 (18.4)
524.6 (21.1)
584.9 (26.6)
541.7 (22.6)
504.3 (19.5)
607 (27.7)
578.7 (25.3)
613.9 (27.6)
554 (19.1)
459.6 (11.1)
481.9 (16.9)
469.6 (15.8)
—
—
—
—
—
—
—
—
—
—
gender congruency
—
gender congruency
—
—
—
—
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)
399
KAZANINA AND PHILLIPS
Principle C and poka conditions
Downloaded by [University Of Maryland] at 05:32 21 October 2011
Principle C
Poka
Region
Gender match
Gender mismatch
Gender match
Gender mismatch
Best fitting model
1
2
472.4 (11.3)
412.8 (9.8)
497.3 (16.7)
417.8 (13.8)
478.1 (14.6)
435.6 (14.9)
485.1 (16.6)
405.6 (9.1)
3
4
5
6
491.3 (19.6)
438.1 (14.6)
589.5 (26.1)
582 (26.4)
456.5 (14.2)
424.6 (11.6)
566.3 (24.2)
639.3 (29.8)
468.5 (17.3)
483.5 (17.5)
430.1 (10.6)
572.5 (23.3)
478.9 (17.4)
487 (17.2)
414.7 (9)
588.9 (27.2)
7
8
9 (2nd subject)
10
11
12
630.3 (24.3)
506.5 (17.1)
496.6 (16)
569.4 (23.1)
513.1 (17.3)
527.9 (16.9)
602.4 (24.9)
515.3 (18.6)
500.8 (15)
584.8 (24.7)
526.1 (19.8)
532.5 (20)
597.2 (28.2)
582.6 (25.4)
559 (18.9)
615 (26.6)
520.7 (18.1)
580.8 (24.4)
587.7 (23.8)
593.3 (23.5)
550.1 (17.3)
570.5 (20.5)
525.3 (19.5)
522.6 (16.6)
13
14 (3rd subject)
15
580.1 (25)
560.2 (19.6)
477.6 (12.9)
578.3 (21.1)
531.1 (18.6)
502.3 (17)
566.8 (20.6)
575.7 (21.8)
526.7 (19.7)
546.2 (21.9)
511.6 (15.7)
472.8 (13.9)
16
17
504.5 (18)
523.8 (18.2)
473.2 (15.7)
512.9 (16.8)
501.4 (16.3)
506 (14.5)
477.1 (15.7)
497.9 (13.7)
—
constraint, gender congruency,
Constraint Gender Congruency
—
—
—
constraint, gender congruency,
Constraint Gender Congruency
—
—
constraint
—
—
constraint, gender congruency,
Constraint Gender Congruency
—
constraint, gender congruency
constraint, gender congruency,
Constraint Gender Congruency
—
—
400
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (2)